download-circle

ClickHouse Will Let You Be Wrong at Scale

October 20, 2026 10 min read in Observability

So I've posted before how I think ClickHouse is great. Its columnar storage is fast, it has amazing compression. I won't bore you with those details, you can see them all here: https://matduggan.com/clickhouse-is-winning-the-observability-wars/

However all technology is great until you try to do something atypical with it.

What I wanted was a realistic torture test: network firewall telemetry, on the order of 2 billion rows a day of mixed-vendor syslog, parsed at ingest into wide typed tables with a 14-day TTL. The dashboards needed to do two things at once in Grafana: fast lookup of individual log lines, and fast visualizations where you can stack ad-hoc filters on top of each other across every panel simultaneously.

I learned a lot of things about ClickHouse. Also some things about myself, not all flattering.

TL;DR

This ended up being significantly harder than I thought it would be, and part of that is that I made a ton of mistakes. Don't email me mean comments. I'm writing this up as a warning to the people coming behind me, and I'll own every mistake as we go.

For the happy OTel path, CH is very easy. Few hundred metrics, couple million logs a day, several thousand traces, not a problem. For the more complex custom route, CH is extremely powerful but it's hard to use.

The theme of the whole project is that this stack fails silently. ClickHouse is the most agreeable coworker you'll ever have because it never says you are wrong, it just lets you be super wrong. A query that looks fine reads 84x more rows than it should. A misconfigured filter just doesn't apply. A projection that never routes still returns correct numbers, slowly. Once I understood that ClickHouse will never tell me that I did something wrong and that I had to figure out I did something wrong, we got along fine.

You have to build verification habits before you build anything else. system.query_log is your friend, your bartender, the lovable dog in a horror movie. This is the only thing in the system that says you messed up.

Here are the lessons I learned. Hopefully they save someone else from going down the same path I did.

Setup

So we're not doing anything weird here.

ClickHouse is a 2 shard x 3 replicas setup underneath the Altinity operator. Every table is a ReplicatedMergeTree wrapped in Distributed tables and Grafana only ever queries the wrapper.
Grafana is the official grafana-clickhouse-datasource plugin, 25 panels and then ad-hoc filters up at the top that apply to every panel at once.
Hard guardrails we cannot break. max_rows_to_read = 1 billion, max_bytes_to_read = 100 GB, max_execution_time = 60s, all in throw mode. I can raise these but the entire point of this process was to operate inside of the CH model for data and try to leave those alone.

Lesson 1: Don't handwrite the filters

My first draft of the dashboard wrote filters the way I'd typically write them in normal SQL, by adding a guard so the query stays valid regardless of what the user picks.

WHERE (empty({{observer}}) OR observer IN {{observer}})

This would end up being quite the stupid decision.

predicate	rows read	time
`observer = 'one-device'` (direct)	3.05M	115 ms
`(empty(...) OR col IN ...)` guard	257.64M	1,134 ms

Eighty-four times the rows, ten times the time, same answer. Why is this so much worse? I'm not entirely sure. My best guess is that the guard keeps primary-key pruning from engaging, so you get a full partition scan with post-filtering. I clearly measured the effect but I'm not exactly sure how to prove the mechanism. EXPLAIN was my only savior here.

The fix was to stop putting in the WHERE-clause filtering by hand and use Grafana ad-hoc filtering. The key learning here is how the plugin does that and why it makes a difference. It does not rewrite your WHERE clause. It appends the filter as a ClickHouse query setting, additional_table_filters, which means it is applied at planning time and prunes exactly like a hand-written query.

form	rows read	time
no filter	256.90M	1,265 ms
hand-written `WHERE observer = '...'`	4.67M	170 ms
ad-hoc filter	4.67M	284 ms

In case you don't know what the grafana ad hoc filter looks like

Lesson 2: Sort keys matter sorta (dad joke, you are welcome)

When I first set up the main table, I did the following:

ORDER BY (observer, action, application, toStartOfHour(Timestamp), Timestamp)

The logic made sense at the time, even if it didn't end up holding up. We're sorting by network device first since of all of these, there are going to be the least number of network devices. But what I didn't really think about is that every panel is time-bound because of course they are, that's how Grafana works. In ClickHouse only the front of the sort key gets the fast lookup path. Obvious in retrospect, but stumped me for a hot minute. I thought I was structuring this by lowest cardinality (device) and then out to most cardinality, which logically made sense.

If all you want is information on the right way to do this, you can find that here: https://clickhouse.com/blog/working-with-time-series-data-and-functions-ClickHouse and https://oneuptime.com/blog/post/2026-03-31-clickhouse-choose-order-by-key/view

What does ORDER BY mean in ClickHouse?

What's important to understand about this index is that it's not the same as like MySQL necessarily. It's the physical layout, the rows are written to disk in sorted order like a phone book sorted by lastname, firstname.

ClickHouse by design doesn't keep a very large "bookmark list", every 8192 rows (called a granule), it writes down the sort-key values at that spot. That bookmark list is the primary key.

So the Fast lookup only works for a prefix of the sort columns:

"All the Smiths" → jump straight to the S section. Fast.
"Everyone named John" → Johns are scattered through every last-name section → you read the whole book. Slow.
Their table also splits data into daily partitions (separate files per day). That's a coarse filter that works no matter what the sort key is (remember this, it becomes important)

Why is timestamp in there twice?

Everyone asked me why I did that, let me explain.

The two entries are not the same thing.

toStartOfHour(Timestamp) — the hour bucket. It rounds 14:37:22 down to 14:00:00. This is the "cutoff"
Timestamp — the exact time.

Within any fixed (observer, action, application) group, sorting by (hour, exact-time) produces exactly the same order as sorting by (exact-time) alone — the hour is just a rounded copy of the same value, so it can never re-order two rows.

So my first design was basically this (observer, action, application, hour, Timestamp).

So time sat at position four, and the most common predicate in the dashboard was aimed at the worst-positioned column. The tail-logs panel with ORDER BY Timestamp DESC LIMIT 1000, a multi-hour window, measured before the guardrails went up, read 291.8M rows, 158 GiB, in 18.9 seconds; with one observer filter, 23.1M rows, 936 ms. The newest rows sit at the end of each device’s block so a newest-N query has to walk every device's block to find the tail.

The fix:

(toStartOfHour(Timestamp), observer, action, application, Timestamp) — puts the bucket first, so an hour range is one contiguous block. All the Smiths together.

Here's what makes this decision consequential: MODIFY ORDER BY can't reorder, prepend, or absorb existing columns. You can only append columns added in the same ALTER. Get the key wrong on day one and the fix is not an ALTER; it's a new table, ~520 GiB of backfill per shard under live ingest, and an atomic swap. Ask me how I know. MODIFY ORDER BY turns out to be less 'modify' and more 'build a second house next door, carry everything over by hand, then swap the mailboxes.

After all those lessons, I ended up running a benchmark that was the kind of beautiful baby benchmark you dream of where it confirms you solved every problem. An identical aggregate query, 156 ms vs 1,690 ms. 10.8x faster! I felt great about that for most of an afternoon, which is how long it took to notice that 88% of the old table's partition had tiered to S3 while the new table sat entirely on local NVMe. We had benchmarked a Ferrari against a tractor and concluded the tractor was slow. Which is true. It is also not the thing we measured. Excellent work.

The same-tier control with both on NVMe produced no noticeable query differences. So was it worth reorganizing? I'm not sure! I think?

This took a long time to do, so triple check this part.

Lesson 3: Projections are trickier than I thought

So when you first run across projections in ClickHouse, they seem pretty great.

A projection is not a separate table you manage. It's an alternative physical ordering of the same data, stored inside the same table. Think of it this way: your table is a book sorted by chapter number. A projection is an index at the back, sorted by topic.

When a query matches your projection, it should just automagically read from the projection instead of the base table. You don't need to change the query, which is great. New data populates the projection on INSERT, which is what you want for this sort of stuff.

Think of it like this:

Every table stores data in parts, which are directories on disk.
When you add a projection, each part directory gets a subdirectory for the projection data with an anonymous table with its own columns, index, granultes.
Projection parts merge at the same time as main table parts, so there's no lag like what you can experience with materialized views.
Projections don't backfill by default but you can backfill them, which is a mistake that I made initially where I added one, nothing was better and I assumed I had messed up.

Projections are great for network firewall data because the data doesn't need to change shape. I just need to optimize my querying of the data for specific usecases.

The biggest misunderstanding about how projections worked is that I made a lot of assumptions about how expression matching worked. I assumed (for no good reason) that expression matching was loose and that CH would do a lot of work to ensure my query was routed to the projection. count() and sum(duration) won't be used by a query asking for avg(duration), which didn't occur to me until I read this: https://obsessiondb.com/blog/clickhouse-projections-complete-guide

Also every filter column must be in the projection's GROUP BY. This means a filter on an absent column → full fallback to raw. This was a fucking nightmare to debug until I figured this out.

WHERE clause	routed?	rows read
(none)	yes	2,022
`toStartOfHour(Timestamp) >= X`	yes	2,022
`toStartOfInterval(Timestamp, INTERVAL 1 HOUR) >= X`	no	12,373,248
`Timestamp >= X` (raw)	no	12,373,248

Here are the basic lessons you must understand to do a large scale project.

The query’s aggregates must be a subset of the projection’s — count() and sum(duration) won’t serve avg(duration).
Every filter column must be in the projection’s GROUP BY — otherwise, silent fallback to raw.
The time predicate must use the projection’s own expression: toStartOfHour(Timestamp) >= X routes; toStartOfInterval(Timestamp, INTERVAL 1 HOUR) >= X — identical values — does not. Same query: 2,022 rows versus 12,373,248. It doesn’t reason about equivalence. It’s a bouncer with a laminated list, and toStartOfInterval is toStartOfHour’s identical twin in a different jacket.

All of this was figured out when I started just obsessively checking query_log. In order to use projections in ClickHouse you have to verify the routing every single time.

Here are my mistakes in no particular order with projections:

Projections don't automatically backfill which is fine but be aware when you run your benchmark and nothing has changed.
You gotta think really hard before you add a projection. It's effectively a copy of the data, so make sure your ORDER BY is correct before you try to layer projections on top of it.
system.projection_parts GROUP BY partition tells you the coverage of your projection. You need this in order to know whether the query has access to enough historical projection data for you to actually materialize the thing you want to show.

Lesson 4: Grafana ClickHouse Plugin has a lot of sharp edges

I like Grafana, I love the ClickHouse Plugin group of maintainers, all good vibes over here. But I misunderstood how much this plugin was going to do for me vs what I had to do myself. This is the part of the review where I say nice things and then hand you a first-aid kit.

A wrong table name in the ad-hoc filter config is silently ignored. An unqualified table key instead of the schema-qualified one returned 79,350,116 rows where the correct key returned 2,314,678. A typo produces a confidently wrong dashboard. The only reliable test: change a filter, confirm the number actually changes, per panel.
Grafana's ad-hoc filter UI offers one operator and one value per key. There is no way to express observer = A OR observer = B through the filter box — a long-standing upstream limitation (grafana/grafana#10918, still open for cross-column boolean OR).
Grafana's time handling and panel model will surprise you in every possible way. Hour-bucket edges are a correctness defect factory.

You will not have any idea that the thing you wrote is wrong until you check it by hand or it blows up. My advice is to configure your dashboards to blow up early and often so you don't find out months later that you messed up the query.

Conclusion

ClickHouse is a good tool, but the more complex the thing you try to do with it, the more sharp edges you'll find. I'm also surprised at the ecosystem lacking stuff I expected. I wanted a linter. I got a side quest to make my own. I had trouble getting a full-fledged real migration utility for rolling out changes with advice about impact and cost and then confirming I can always roll back. There are still real gaps in this ecosystem people should be aware of.

As always if I got it wrong (or you know something I don't), email me.

Make tmux the OS

October 02, 2026 19 min read

I recently watched the talk by Scott Jenson titled "Are we really going to use the same Desktop UX forever?" https://www.youtube.com/watch?v=V7AfAcQwLW0&t=445s. He's a great presenter, really articulate and concise. The kind of speaker that you'd gladly listen to for 3+ hours if given the chance. Which for a talk about window management is quite the compliment.

The overall point of the talk was "Apple and Microsoft aren't going to innovate anymore in the desktop OS space, so it is up to us all to decide what the desktop of the future is going to look like". I would argue that the actual situation to more nuanced than that, they're trying ideas they are just extremely conservative. As Jenson points out, a lot of what we treat as "how computers work" was a workaround for hardware we stopped having years ago. So why are we still accepting those tradeoffs?

To be clear, I'm not a UI/UX designer. I don't know what I'm doing and I lack the skill to make this. I'm making this mostly as a thought exercise and hopefully a prompt to get other people to think about this same problem. I have, however, suffered through others bad choices, which turns out to be most of the qualifications required.

TL;DR: The thing I want is "tmux is the OS", but I want tmux where a normal person could use it. I use it all the time, it's great, but is there a way to take the amazing experience of a scrollable, session-persisted, detachable, task-oriented system to normal people?

What are the high-level problems with windows management in 2026?

I think you can break the problem down to the following components:

My screen real estate is constantly changing and I have a lot more of it.

I am going from my laptop screen to my desktop monitor back to the laptop screen about 6 times a day. When I'm on my big monitor or multiple monitors, overlapping windows don't make sense because I have too much space as it is.

First, one big monitor is a a different experience entirely compared to 2 smaller monitors. I don't use the two monitors as one large unbroken screen, I tend to use one of them for "less important static content" like a ToDo list or my work chat tool and my email and then the "primary" monitor for my terminal/tmux/browser which is how I do all my actual work. But windows "belong" to one or they "belong" to another. Grudin found in 2001 that people use a second monitor exactly this way, where focal work on one, peripheral glances on the other. Twenty-five years ago. The OS still doesn't know which monitor is doing one role and which the other.
And when I pull the laptop out of the dock, everything collapses onto one screen and I manually drag and resize windows to claw back enough real estate to keep working. I don't know how much time this actually wastes for people, but it feels like it wastes a bunch. The OS treats a display change as a surprise, instead of as a scheduled event that happens six times a day.

Different people use windows totally differently.

We have three distinct groups of people that we are trying to design around, while most modern OS optimizes only for the first use case.

Some people are window maximizers, where every window takes up most of the screen and then they use Command + Tab or the Dock or something else to switch between the bigger windows. The value of more screen real estate is mostly that this one big window can be even bigger.
Some people are near maximizers, where one window takes up almost all of the screen real estate, but they'll have one or more smaller windows where they glance at things like status or chat or whatever.
Finally there are people who carefully coordinate all of the windows on their screen.
Almost everyone has "private windows" and "public windows". You are fine with your public windows being lined up and persisting, but you want to hide specific information in other windows from people walking by. The people I complain about are also the people I present quarterly numbers to. I never remember that when I connect my laptop to a display they're going to see my entire screen.
I assumed this was just me with private vs public. But it was measured in 2004, in "Revisiting Display Space Management: Understanding Current Practice to Inform Next-generation Design." Same three types, same public/private split, twenty years ago. See how absolutely none of this is a new idea? Link.

Browsers are mini operating systems.

In 2026 everybody quietly agrees that most of your software runs in a browser tab. These web applications will cover a wide variety of use-cases that have historically been owned by local applications. My desktop OS treats a browser just like a normal application, when in fact it is closer to a virtualized OS running inside of my host OS.

Tabs inside of a browser and windows of that browser contain the same level of complexity as my other applications.
Tabs are associated with streams of work alongside my conventional applications. I'm writing Terraform in Vim in my terminal while referencing the Terraform docs for that provider. But the relationship between tabs and work is messy: the same docs tab pulls duty while I write the code and again while I write the ticket update. Whether a tab should be allowed to belong to two tasks at once, or whether something cheaper is going on, is the exact spot where I break with the research. I'll come back to it once you've met WindowScape.

The concept of "filesystem" is an increasingly weak concept.

My notes in Apple Notes belong to Apple Notes, not my filesystem. My texts live in iMessage's database, not my Documents folder. Teams and Slack content lives inside those applications unless I manually "bring it out," and when I do, I'm making a copy. Firefox will happily show me a PDF, but the PDF doesn't live with macOS unless I take an action to make it so.

Now a lot of engineers are going to read that and think "well you cannot force all applications to use the same storage system for all of their files, are you a lunatic?!?". I'm not suggesting that, in fact the weak filesystem might be a perk. There is a CHI paper from 2004 called, and I am not making this up, "Stuff goes into the computer and doesn't come out." That was 2004 and we have the exact same problem with no real solution. Link.
Here's why this is a windowing problem. Since Windows 95 and System 7 (which is basically as old as my memory goes back to), the machine has worked as a chain: something writes a file, the user opens an app on it, the app writes it back, the file gets sent to someone else, repeat. Every link in that chain assumed the file lived somewhere the OS could see. Every link is now broken in all the modern OS.

What have people already tried?

So I'm not the first person to see this problem. Some of it got solved, but in a different direction than I want. Some of it got solved, but not for normal people.

I started with the document that I kept seeing everyone else cite to. Henderson, D. Austin and Stuart K. Card. “Rooms: the use of multiple virtual workspaces to reduce space contention in a window-based graphical user interface.” ACM Transactions on Graphics (TOG) 5 (1986): 211 - 243. Interesting that even back in the 80s there was a pretty clear understanding that the current system for managing windows wasn't very good. The design they were talking about looked something like this, which is pretty advanced compared to where we are.

Henderson and Card measured window use the way operating systems people measured memory. The screen is RAM. A closed window is a page swapped out to disk. And windows, like memory pages, don't get touched at random: you sit inside a small set of them, two to ten, and that set is the task. Programs spend about 98% of their time inside one of these sets, and roughly half the cost of running happens during the 2% of time spent switching between them. Rooms' whole design was preloading the next set before you ask. They described this as reducing "knowledge faulting in the user," which is the best phrase in the literature and I intend to use it until someone stops me. Just try it out in a corporate meeting: "we need to reduce knowledge faulting in the user".

One of the more interesting side papers I found was "No Task Left Behind? Examining the Nature of Fragmented Work". Link. Mostly because it confirmed something I've long suspected, which is the single task for a long time focus on modern widowing systems isn't actually how people work. The title of the companion paper is a real participant quote: "Constant, constant, multi-tasking craziness." People were juggling around ten "working spheres" a day, minutes at a time.

The closest to what I wanted is from the paper WindowScape: A Task Oriented Window Manager. Link.

WindowScape dropped explicit grouping entirely so now every time you changed the arrangement, it took a photograph, and you went back into photographs instead of filling containers. I love their one-line diagnosis of every system before them: "requiring windows to be in a single group forces users to decide ahead of time where a new window belongs." The photograph metaphor also cracks a problem I'll get to with browser tabs: one window can appear in many photos, because, as they put it, "people understand that there can be several photos of an object with there being only one underlying object." The catch is that the photos evaporated. That's the gap that I think you'd want to solve.

For the first problem, we have more or less already solved for overlapping windows. A tiling window manager ensures that you are maximizing your screen real estate in such a way that you can switch between different layouts with no need to manually modify the windows. There are basically 2 problems with tiling window managers as they exist now.

They're way too hard to use. Like an order of magnitude too hard for normal people to use. Basically if you need to start a sentence with "just open up the configuration file" shut it down the thing is over. The average tiling window manager tutorial asks you to clone a repository before it asks you to open a window.
We need a scrollable tiling window manager. So basically I should be able to set up specific window configurations over here, leave it alone, then scroll to the right and do something else different, like shoving the mess into the back of a drawer. It should be infinite space to work without needing to subdivide the windows into different macOS Spaces.
The scrolling also solves the private vs public window problem. I put my private stuff on the far left and then my public stuff on the right. If I want to look at the private stuff, scroll to the left as far as it will go.
Thankfully this already exists with Niri: https://github.com/niri-wm/niri. It just has to be easier to use. But the tough design elements more or less already work.

Browsers as mini-operating systems: people have tried to solve this, but in the wrong direction. They made the browser more of an operating system instead of making its contents first-class citizens of the one you already have.

The best two examples of this are the Arc browser and Chrome OS. But I think Arc is actually the more interesting of the two experiments. Their first innovation was to break the idea of tabs at the top and instead move it to be a sidebar model.

They also basically "took over" the concept of windows from the OS.

The diagrams above and a good write-up of Arc is available here: https://blakecrosley.com/guides/design/arc.

All of this makes sense from the perspective of "the browser is now the operating system", but I think this is a fundamentally flawed idea. If a web application is operating as an application, it should be its own window. If it is complimentary to another application, it should be a window associated with another application and be allowed to contain many tabs, to reflect the idea of the browser as the portal for all research and lookup.

I was shocked to go through the historical progress of people trying to solve this problem. I'm not the first person to try to solve this problem, I might not even be the 10,000th person.

Graveyard of Attempted Solutions

Windows Timeline (2017–2019): All of your activity across all of your apps sorted and organized for you.

Windows Sets (2018–19, canceled): apps and webpages in shared tabbed sets. Note what this actually was: the operating system grouping apps and web content into tasks, which makes it the closest thing anyone has ever shipped to what I'm about to propose.

macOS Stage Manager (2022): automatic task-based window grouping, widely ignored. The reason I think Stage Manager didn't scratch this itch for people is that its super designed around the iPad style flow of "there is one thing you are using at a time and you need to be able to quickly switch between them". On iPad the model is "one thing at a time, switch fast." On desktop, two or three things have to work together. It compromised toward the iPad and hit for neither.

KDE Activities: Everything I'm writing here is old news for KDE. They've been doing task-scoped desktop state for over a decade. Honestly this does most of what I'm writing about, it's just hyper manual and requires you to manage and set it all up. Also I love the KDE design docs, amazing stuff, worth reading. https://community.kde.org/Get_Involved/design

PWA install: already gives web apps their own window and no browser chrome, which is clunky but at least makes them "real" applications.

So clearly we understand there's a problem. Why haven't these taken off like wildfire? I think there's a couple of different problems here.

Timeline needed apps to opt in to reporting activity, and most never did. It also logged everything, which read as creepy (which is a problem that my idea would have too), and the UI surfaced Edge features nobody wanted, so it felt like a browser push more than a feature. Sets died somewhere between internal strategy and app-compatibility chaos where the interesting question is why nobody demanded it loudly enough to save it. Stage Manager tried to be universal and pleased nobody. KDE Activities is opt-in, and opt-in means the people who need it most never configure it.

The pattern in the graveyard: the ideas aren't wrong, they're just not defaults, or they're not done at the OS level where they can see all apps. Everything I want has to be structural.

Filesystems as a weak abstraction.

Operating systems have tried to solve this, but because there's no requirement that you use the user OS filesystem to store documents, there's no consistency. MacOS has recent files, but "recent" doesn't mean anything (these are not the most recent files I have downloaded to my computer). As far as I can tell this functionality is completely broken or implemented in a way that makes zero sense to me.

I have opened, downloaded and created dozens of documents since September 23rd (I'm writing this on September 28th). I don't have a single fucking clue how MacOS populates this window. Maybe its broken. Maybe its working as intended through some criteria I don't understand.

Maybe it's personal.

But the idea is here. The ideal would be "across my entire computer and applications, what are the recent files I have interacted with" and expand that out to include emails and Teams/Slacks and everything. I should be able to see everything going on with my machine, search through it and not care if its stored inside of Slack or iCloud or whatever. Single pane of glass.

Microsoft tried it, people didn't like it, but I think the concept makes sense.

What might a solution look like?

So what has changed? Why might we be able to crack this problem now when before it was too complicated? This is where I think a local LLM might make sense, if you can figure out a way to do it where it doesn't cause more problems than it solves.

So what are we looking at here. The concept would be organizing it around the idea of tasks. When you define a task, this allows you to organize all the windows together, including specific browser tabs detached and associated with the task, not with the concept of "browser". You still get get the flexibility of defining glanceables that exist outside of the strict tiling view.

The basic window flow would look like this:

The desktop is one infinite canvas; each physical display is a viewport onto it.

You would end up with the following "transition contract" to handle my initial problem of "what about many monitors/new monitors".

Never Change items: horizontal order of windows, scroll position, input focus, task membership.
Reflows Deterministically: column widths, like a responsive layout. When the shelf narrows, the books stay in order, the rightmost ones fall off the edge of the viewport into the scroll region
Is state, replayed on redock: viewport aims. Two monitors means two viewports aimed at two regions with one at your focal task, one at the glanceables region. Undock: second viewport disappears; its region is one scroll gesture away.Redock: it re-aims where it was.

The OS finally learns which monitor is focal. The display receiving keystrokes ~90% of the time is focal; windows that stay visible but rarely receive focus are glanceables. That's inferable from focus telemetry and requires no eye tracking or config. And when the laptop screen is small, glanceables demote to a thin strip rather than full tiles which is, note, exactly what tmux already does.

One substrate that covers our three user types. Remember the maximizers, near-maximizers, and coordinators? On the canvas they're just column counts. A maximizer is one full-width column. A near-maximizer is one column plus the glance strip. A coordinator is N columns. Nobody gets forced into anything. This is why the design can be a default where KDE Activities was a preference. It's because the substrate doesn't impose a style, it just stops punishing whichever style you already have.

Privacy becomes a first-class flag, driven by machine state. Private is a property of a window or tab, not a region of the screen. Private content renders occluded unless the machine can vouch for safety. The rule I landed on, at the cost of my favorite bad idea (story below): derive privacy state from machine-observable facts like connected displays, or active capture sessions and never from inferred human states like attention or idleness.

Connect to a novel display and it gets a "ready to present" screen by default. You explicitly aim it: this task, this window, or extend the canvas. Known displays skip the dance. Screen sharing is the same event with no cable. You don't need the model to have discipline if the door won't open, and you don't need the user to have memory if the default can't leak.

None of this is exotic. Apple's Keynote's presenter view has been shipping for twenty years plus with slides on the projector, notes on your screen. Apple never promoted the pattern to the OS even though it makes obvious sense. I'm asking for presenter view as a desktop primitive.

Browser tabs detach into real windows and join whatever task they belong to. The "browser" stops being a place windows live.

This all makes sense until you get to "how do you organize this stuff around tasks". I think for more expert users they're going to be able to do this stuff, but part of the problem is that we don't want them to ever have to drop back down into a configuration file. A mouse and dragging stuff around is too clunky for how this would work. Ideally I should be able to ask something that has access to what information is contained inside of each one of these browser tabs and window and help me organize it in some logical flow.

That's where the local LLM comes in.

I'm calling it a "quake overlay" because I'm old and the idea of a text input that drops in from the top with a universal shortcut has always been the quake terminal dropdown to me. Younger people know the same pattern from the Discord overlay, which I have decided not to be mad about. However in the previous diagram its shown as a more friendly "Ask" box.

But the basic flow would be that you ask the LLM to assist you with organizing, it would show you what it's thinking by querying the information through a constrained MCP server with a set verb list and then gives you a preview of what the layout is going to look like. The model never touches the window manager directly. It proposes and you press y.

People rarely pick up completely novel tasks: a graphic designer spends their life in "open files from network storage, make changes, save output, paste into chat, repeat." People rarely sit down at novel display contexts: laptop-only, the desk, the conference room. People rarely attach novel displays. Novelty is rare everywhere, so the expensive one-time machinery of classify, propose, preview, confirm only runs rarely, and everything in between is deterministic replay. You organize a limited series of tasks once and reuse them forever. Check the tickets, open Vim and a browser, work the ticket, write the update, open the PR, post it for review, next ticket. You could make the config a nightmare of JSON and stop caring, because the only intended reader is a model that doesn't mind. The config file stops being an interface.

The biggest issue here would be spatial memory. People understand where things are in relationship to each other on their computer and they don't like it if they are changed. Think of it like "if I came and messed up your physical desk by moving stuff around and adjusting your chair". So I suspect you would need to enforce a strict "append, don't replace" model.

There's a name for this actually. Kirsh and Maglio call it epistemic action, where Tetris players rotate pieces more than the game requires, because rotating is how they think about the piece. Link Your window layout is thought in progress. A helper that "optimizes" it mid-thought is interrupting you.

This is a problem I encountered with just trying this idea out though. Modern LLMs are actually not very good at "touch this stuff, never touch that stuff". So unclear if this is a realistic idea or not. One boring fix: the LLM proposes, dumb deterministic code executes, and the executor refuses to touch anything pinned. You don't need the model to have discipline if the door won't open.

You'd still need a high level UI element for users and this is where I think the more inclusive concept of filesystem could come in.

Tasks become roughly comparable to Directories now. But everything flows from the initial high level concept of "Tasks" and then flows out to individual things. One application is never the task. It's "some chat window, a browser, maybe Preview." Can we try to build it?

So my hope was that here I would be able to post a Linux desktop running an example. But as it turns out (unsurprisingly) it is insanely complicated to make something like this run at all. Some of the underlying ideas work reasonably well (global terminal shortcut dropdown from the top, scroll-able tiles), but it's still pretty clunky. I'm gonna keep working on it and see if I can get something as a demo running, so if you are interested either add me to an RSS reader or just....wait I guess.

In the interest of full disclosure I think this is gonna take many weekends of work to even get a functional demo running. So I'll do my best, but be patient.

Reasons this design sucks

So in attempting to get this up and running in a Linux VM, I immediately ran into serious logical issues with my design. I think failures are often more interesting to read about than successes, so let's talk about them.

The privacy-minded design runs on surveillance-shaped data. There's an irony I sat with for a while. The design promises the OS will finally know which monitor is focal which is the thing Grudin measured twenty-five years ago and the mechanism is input-focus telemetry. Watch which display gets the keystrokes. Watch which windows never receive focus. Store that over time. The privacy-first window manager wants to know more about what you're doing than your current OS does. Keep it local, keep it ephemeral but it's still a lot of behavioral data to drive this system. Just the observation that every mechanism in this post is hungry for data that has historically derailed ideas like this.
My favorite privacy feature was a joke. The original plan was elegant: private stuff far left, public to the right, and when you go idle the viewport drifts back to public. Like a friend changing the subject for you. Then I ran it, and the first thing I learned is that reading is an idle state. You stop typing to read the thing, and the system starts scrolling the thing away from you. Meanwhile, a person walking by sees everything, because a "private" window on a wide monitor isn't hidden it's just further left. The feature hides content from its owner and shows it to the threat. Privacy by occlusion needs to know when a threat exists, and the threat is a pedestrian, which is undetectable.
Nothing in this design ever ages out. The canvas is infinite, the rule is append-only, spatial memory is sacred so the mess grows monotonically. I have solved the electronic messy desk by buying it an infinite desk. When I first read WindowScape, I called their evaporating photographs "the gap to fix." I've changed my mind because evaporation was quietly doing the job of a garbage collector. Something needs to go dormant without moving but deciding what goes dormant is a reaping decision, and reaping is exactly what append-only forbids. It didn't take long for the infinite desktop to feel overwhelming when I tried it.

Fun experiment

I don't know if this underlying idea is a good idea, but it is liberating to stop waiting for MacOS and Windows to do something better or interesting in this space and decide to try it yourself. The graveyard up there is full of good ideas that died of distribution. We're overdue for a radical experiment that ships as a default.

The most eye-opening part of this entire experience was how much the original overlapping window design was a hack to get around fundamental hardware limitations and how soon after it was launched did people clearly see the problems. I'm surprised how often that happens in technology, where you see a problem, search for academic papers about the problem and find just a massive wealth of information from people saying "oh yeah this is 100% a problem and one we should solve soon". Maybe this post inspires someone smarter than me to solve it finally.

I’m Tired of Being on the Network

September 25, 2026 10 min read

You can tell how tired someone is of the always-online life pretty easily. When they get home, watch the distance between the phone and the body. The people who love being connected never let the thing leave them be it pocket, palm, face-down next to the fork. Everyone else sets it down the way you drop a heavy backpack after a long day of walking. My wife does not set her phone down so much as launch it. From her hand to the far corner of the room, the phone travels from indispensable to litter in under a second.

For my part, I've come to resent the smartphone, which is an embarrassing thing to admit, because I used to be a believer. I researched every new model. I switched between iOS and Android just to taste each new future as it arrived. I lined up for the first iPhone at 3 a.m., stood in the dark to get a glimpse of the thing itself. I remember being genuinely moved when the Apple Store employees clapped for me. In hindsight this is mortifying, roughly on par with the year I dyed my hair blonde and shopped at Hot Topic. We used to have passionate arguments about whether iOS or Android was the future. Sitting here right now, I cannot tell you which model of iPhone is in my pocket. I can't name a single reason I'd upgrade, except that eventually the battery dies, and I don't feel like swapping it myself. Again.

It was a dream, for a while: internet everywhere, always. Star Trek made real. I could send an email from a country road. I could read a webpage from a bathroom, which, fine, is where a lot of it belonged. My first computer had no networking at all, and then: dial-up, ethernet, wireless, the phone in my pocket. Each step felt like proof that the future arrived on schedule. These were prized possessions. You cherished them. You had feelings about the merits of one versus another.

Now the smartphone is a tax. I get to choose my tax collector be it Apple, Samsung, Google, Nothing, Fairphone, but I don't get to choose whether to pay. In Denmark I'm expected to carry the Aula app for school, the Rejsekort app for trains, MitID to prove to the state that I am me, and e-Boks for official messages from the government and its authorized friends. I need the thing at the airport, at the hotel, at the gym. All of this, I'm told, is a choice. I "choose" to take part in the digital economy.

It isn't a choice in any meaningful sense. It's compliance. It's the one physical counter open Tuesdays from 10 to 12, and the ticket machine that no longer takes cash. And the offline option exists the way the wheelchair ramp exists at certain train stations: installed after the fact, satisfying a regulation, pitched at an angle no wheelchair on earth can climb. We are all expected to pretend grandma can make it up, and to act only mildly surprised when we find her at the bottom again.

So every few years I buy the identical slab of metal and glass. I admire the color for a moment, because color is the only decision left to make. The last honest pleasure in a new phone is peeling the plastic film off the screen after which the box contains nothing of interest. I put the slab in a case so durable I never see the color again, restore from backup, and everything the old slab knew migrates, whole and continuous, into the new one. A soul transfer, over Wi-Fi, in ninety seconds. I feel nothing for the old slab, despite years of daily intimacy. I hand it off or sell it and never think of it again.

I understood something had gone wrong in me the first time I dropped the phone and didn't mind. I was standing in my living room, mid-negotiation of a play date over WhatsApp, when an email arrived from my mother and a calendar alert reminded me I'd never scheduled the dog's vet appointment. It was as if the phone simply left my hands on its own. It hit the floor with that specific thonk of expensive object meeting laminate. When I picked it up and saw that the screen had survived, I was a tiny bit sad.

After that, I couldn't stop seeing everything I hated about the thing. It was like the last months of a relationship, when everything the other person does induces fury. Why am I out of storage? I don't control the size of the apps that install themselves, or whether my photos are kept at full resolution or quietly compressed into nothing. I'm not allowed to control any part of the experience, so I don't consider the disk space my fucking problem.

Apple pings me, yet again, about my three complimentary months of Apple Music. I open the App Store and find an endless sea of garbage with ads floating on top of the garbage, developers screaming for attention on the burning trash river that is the iOS App Store in 2026. My maps have ads. My apps have ads. Many of the ads are for installing more apps. My podcast app shows me an ad for a podcast about digital detox.

I get on my bike and start a podcast. The first thing I hear, at maximum volume, is a Danish advertisement for a new kind of sandwich bread. Then I learn about the grand opening of an electronics store at a mall near me. As I round the corner and join the mob of Danish cyclists, furiously ringing their bells at tourists, the hosts remind me I can subscribe to the premium edition of the podcast. I long for the days of driving through the night with AM radio, where at least the ads were interesting.

I have come to actively hate this rectangle of annoyances. It is a rectangle I carry from room to room and charge every night so it can go on notifying me, on behalf of applications I was forced to install, because somebody's OKR that quarter was "increase app installs by 15 percent." I don't need to know about the AliExpress sale. There is always an AliExpress sale.

To keep from subconsciously smashing the current slab and having to buy its successor, I started leaving the phone by the front door, next to where I stack the garbage. But the fatigue found me anyway. I'd sit down to play something and get hit with a dozen download prompts. The Steam Deck is always downloading something, or trying to. Streaming has ads now, or else the show has quietly gone missing. My wife and I say insane things to each other, like, "Which app had the show from last week?" HBO counts it off — "ad one of seven" — then detonates sports-betting commercials at maximum volume, until whatever fragile interest I had in the show dies of exposure.

There is an incredible sameness to all of it. Everyone online speaks in the same broken loop of the same ten phrases. Every streaming show (except, weirdly, Apple TV+, and who called that) is written as though I'm the stupidest person alive, plot beats arriving on schedule, like a waiter screaming the order at a customer with a head injury who has already nodded twice. Emily in Paris is on: the story of an American woman trapped in a Parisian time loop. She's with the French chef. Now she isn't. Now she is again. Different dress. Compared to what's arrived since, poor Emily is Shakespeare.

For short-form, TikTok offers a stranger's very confident opinion of an article they did not read. Instagram offers puppies and softcore pornography. For the long-form connoisseur there's Twitch, the premier cyber-dystopian hellscape, where a woman in a bathing suit films herself in a windowless room for twelve hours a day, or a man with nineteen subscribers sits in a sad home office on patchy carpet pleading, "Guys, come hang out with me!" I lingered once, out of a guilt I still can't locate, and he read my username out loud and welcomed me by name. I wanted to scream, not at him, but at the sheer horror of the situation.

The snake has begun eating itself. Netflix has shows about TikTok stars running their social-media empires. HBO has a murder documentary where most of the commentary is social-media clips of people theorizing about the murder. It's a closed loop. Soon we'll produce the documentary while the original event is still happening.

Whatever hope I had for the internet is long dead, and I've made a shaky peace with most of what killed it. We persist anyway because I can't retire and wouldn't have anything to do if I did.

How do you recharge your batteries in hell?

For me: a bare circuit board. I still like playing with tech, I just need something contained. I need something where I can just engage with something that doesn't involve a million hooks and attempts to get me to do something else.

I'd heard the MiSTer was the best way to play old video games. There are a million ways to play old video games now, and that's sort of the problem. Most "retro" devices come with fourteen million games preloaded, which means you play none of them. My whole session becomes scrolling an infinite menu of things I half-remember from the backs of comic books, dipping in and out, retaining nothing. What I wanted was fewer choices. Something that stays offline.

So I ordered one. Setup was easy: flash an SD card, run an update script, copy over the games. Then I pulled the ethernet cable out, and it kept working exactly as before which is both the entire point and, in 2026, a minor miracle. There is something deeply liberating about owning one object that does not attempt to walk me through an onboarding flow. No EULA. No cookie consent. From my television to my goddamn refrigerator, everything else in my life is clawing its way onto the internet. The MiSTer just sits there and does the job.

The refinement that made it click is TapTo or Zaparoo, depending on how old the website you're reading is. It's an NFC card system that allow for basically a physical shortcut to the game. Tap the card, the game loads. It sounds like a gimmick but it is actually the secret to why the whole thing works. It's the difference between scrolling a menu and picking a cartridge off a shelf, and it finally scratched the itch. (Time Extension reviewed it here: [link]. If you want the full rundown on what a MiSTer is and how it works, RetroRGB is the canonical source: [link].)

The short version of what the MiSTer does: software emulation is a very good impressionist doing an SNES. FPGA is reincarnation with the original chip logic rebuilt, cycle-accurate, no frame skip, able to drive a CRT if you happened to keep one. Is it objectively better than emulation? Who cares. Every slowdown, every piece of weirdness, belongs to the original game and not to some setting I need to go find. I do not open menus.

The end result looks like this (not my actual unit — I dropped it and cracked the plastic so: imagine this, but worse):

The first game I loaded was Crash Bandicoot for the PlayStation. When I played it as a kid, it was a technical marvel. It is hard to overstate how Crash looked and felt, coming from the Genesis and the NES. PS1 games in general have a soothing grammar, because they assume nothing. They know you rented this from a Blockbuster on a Friday night, that the manual is missing, and that you need to understand the game inside of thirty seconds.

Each Crash level introduces exactly one new thing, shows it to you up front in a low-stakes loop, and lets you figure it out. It starts easy. Then the mechanics stack, and stack, until the jumps demand something close to pixel-perfect, and I notice I'm leaning forward on the couch, fully dialed in to the triangle-headed marsupial.

It puts me back in the arcade at the mall a town over from my house. My dad would drop me off at the Defiance mall, which had a small arcade near the movie theater, across from the Chinese food stand in the food court. We'd line up in the dark for a shot at one of the Street Fighter II machines, quarters stacked on the bezel to call the next game. The carpet was always weirdly sticky. The joysticks were still sweaty from the last failed challenger. The mechanics of that era were simple, but you had to execute them perfectly, in public.

After a few hours I take a break. I pause the game and leave the box running, and hours later I come back to find everything exactly where I left it. No session expired. Nothing has decided I need to log in again. It just keeps humming, the little fan I installed purring away. It exists to keep doing the thing I asked it to do, and it has no other ambitions.

When we first got computers, this is simply what they were. By default they did nothing. They sat there and waited for you to have an idea. My old Macintosh LC II would basically wait around for inspiration to strike, or for me to waste an afternoon opening and closing folders, because software was expensive and folder-clicking was free. They felt like tools, not toys, and (ironically) the reason we knew how to do anything on them was that we spent hours poking around System 7, figuring out where everything lived. They crashed. Often, if we're being honest. But nothing in that machine wanted anything from me.

Is the MiSTer perfect? No. It's a board engineers test things on, and the software is flawed in exactly the ways you'd expect software like this to be flawed. The UI is clunky, pairing your first controller sorta sucks, and at no point has anyone optimized my experience. But it exists not as the physical manifestation of some roadmap I'll be subjected to later, but as a snapshot of a thing that exists on its own terms. Like the LC II, it's a thing designed to do a thing. It sits there with wires dangling out of it, doing exactly that.

It asks nothing of me. It knows nothing about me.

I Am a Flat-Rate Monthly Responsibility Service

September 11, 2026 7 min read in Personal

Most people, when asked why they do what they do, lie. This isn’t because they’re malicious but it’s because the honest justification for a career is rarely noble. It’s usually a combination of a decent paycheck, tolerable hours, and whatever neurosis you were nursing at that particular juncture of your life.

Take my friend the ER doctor in Chicago. He’ll tell you he went into medicine to help people. The truth? He became an ER doctor because he’s the kind of person who cannot do the same thing twice. He requires the chaos and strangeness of a city emergency room the way a 90s sitcom dad needs a beer and to watch the game. Helping people is a delightful side-effect of his pathological need to never sit still.

I fell in love with tech work the same way a 13-year-old falls in love with anything. It was the late-90s, and I was in a small-town computer shop that smelled of burnt ozone, cheap carpet, and cups of noodles. I didn’t love technology as its own thing. I didn’t believe it was a transformative tool for the human mind. I liked fixing desktop computers because I got a dopamine hit from solving the riddle. Why is the video not showing up? Why can't I do this in the UI versus that? These were problems people were grateful to have solved, but honestly, their appreciation meant nothing to me. I was fascinated by the puzzle until the moment it was solved, at which point the entire affair ceased to matter.

Technical problems, if you have the sickness, are among the best. They can be insanely complicated, but really, nobody dies if you fail. If you poison a soup, someone goes to the hospital. If you bork a database, it's just a bad night. Since everything in the entire stack is knowable at some level and designed by humans, intuition often gets you further than rigorous logic. They’re immensely satisfying because they require hours reading documentation or source code i.e. there is a toll to be paid in the mines before you see the diamond. But the underlying solution is usually quite simple, easy to explain, and fast to test. You confirm your hypothesis, ship the solution with your name on it, and wait for the next fix.

If you like to test and iterate on your theories as you work, software is a dream. I don’t need a lab, or equipment, or really anything other than any laptop ever made. It’s like being a scientist if the only thing you needed was one test tube, one beaker, and some loose mercury rolling around on your desk.

The driver for me going further and further down the stack was that you don’t get a rush for solving the same problem twice. The first time I diagnosed a blown capacitor on a videocard, there’s a rush. The tenth time, it’s just a job. Fixing problems in a GUI was good for a while, but eventually, you want harder problems that exist exclusively inside of CLIs. Debugging closed-source software is boring because you capture the logs and produce a bug report like a good little citizen. Open-source software lets you go right down to the metal.

Then you get to the next level, which is: What if I can solve the problem by making a better version of the thing? This is the zenith. Now you are perhaps the most expert in the world on fixing this thing. You made it. The call is coming from inside the house. Every problem is a new problem, often one you never predicted. It’s nested endless puzzles stacked on top of each other. Then your job is sitting quietly in the dark, headphones on, listening to music and solving puzzles, which, sweet fucking Christ, is paradise on Earth.

If fortune smiles on you, maybe you don’t have to manage actual humans to keep getting promoted. Maybe you get to just wander around like a modern-day ronin and solve problems in other stacks for the remainder of your career.

I’ve met other tech workers with the same sickness as me, but we’re a minority. In my experience, most people who work in tech can be sorted into very specific buckets:

"I'm here because I love technology."
Sometimes they are obsessed with OSS and how it’s going to save the world. Sometimes they’re programming language nerds who delight in solving problems with clever quirks of their chosen syntax. These are the lunatics who say things like, "Writing my own programming language was a lot of fun."

"I like high visibility, low stakes work."
Some people absolutely get off on working for Company Name Everyone Knows, maintaining a product people rely on daily. If it stopped working, it would be a serious problem, but again, nobody dies. Maintenance and keeping the lights on is the name of the game. God bless them.

"Money is here, so I am here."
No shame at all. Some people needed a job, tech skills were hiring, so they’re here. They do just fine work. Honestly, they tend to get stuck in the weeds far less than the "I love technology" people, who pick weird esoteric ideological hills to die on all the time. This is the largest group by a lot now. They are the mercenaries.

"I'm here for a hard time, not a long time."
This is me. I want the worst problems in fast succession, which I will complain about, but if I solve them all and the job becomes easy, I will get bored and leave. I’m like the worst kind of restaurant customer. I order everything on the menu, eat it all, complain that I’m too full, and then when you don't change the menu next time, I never come again.

Now, the LLMs have arrived, like a fleet of Sysco food-service trucks backed up to the kitchen door, delivering pre-chopped onions and powdered hollandaise. They came for the "I love technology" people first. There’s no prestige in being an active contributor to an open-source community anymore because you are just a meat proxy for the Borg. Make sure your docs are formatted for them!

Now in LLMs defense, it's not hard to make the "I love technology" people upset. These are people who I've seen quit jobs over framework debates. I once watched a senior engineer get into a yelling match with another over whether we should mandate emacs as the official text editor of the team. Their desire for the most pure thing often overrides any ability to do cost benefit analysis. They're closer to musicians than factory workers.

However now, it seems they are coming for me.

So here’s the issue. LLMs make programming much easier. Because it is easier to make a thing, you care less about fixing the existing thing. There is no hit from solving a problem, because who gives a shit? Nobody is impressed that you sat there, studied the docs, and read the source code, because an agent will do that work for you while you eat a slop bowl on a Teams call.

I used to get really excited when someone made a new Mac application. Typically, these were deep passion projects by people who loved the ecosystem. Indie Mac developers are the best part of owning a Mac. SoundSource, BBEdit, Things, Dropover—the list goes on. It was craft. Now I am flooded with LLM-generated Mac apps and I just don't care. All it means is someone paid Apple the $100 developer fee and gave OpenAI $250 to stitch together a UI, then copy-pasted error messages into the prompt until the thing finally launched. Nothing was mastered. Their passion for solving the problem didn't push them through the hard part. They had an errant thought, manifested it into reality, and are now showing it to me. It's exactly like a coworker cornering you in the breakroom to tell you about a dream they had. You have to nod, but you are dead inside.

You see the same thing in the CLI space. How it worked before is someone would decide they needed to solve a problem. They would put in a ton of work, then launch the thing. You'd find it on Hacker News or wherever. There was a vested interest in working together on this thing because yes maybe it didn't do what you wanted perfectly but it was way closer than starting from scratch. A community is formed about taking this work and then extending it in the directions that allowed it to maximize its usefulness as a tool.

Now why bother? If this CLI tool doesn't do what I want exactly, I'll just wait the....72 hours until someone has the robot make a new one. It's like gaming before Steam and after Steam. If a game doesn't hit exactly for me now I just refund it before the trial period is over. There's no reason to stick around and follow its progress. There will be another Metroidvania in the next 36 hours. Important to note, games aren't better now that there is a firehose of them, they're more niche and the prices have been driven down. I don't get the impression that these CLI tools are more stable or trending up in quality, there are just a lot more of them, more than a person could reasonably try out.

I tolerated a lot of the ridiculous bullshit of the tech industry because, at the end of the day, I got to return to my endless series of puzzles. Stand-ups. Meetings with investors. Retros. Kanban boards. Agile vs Waterfall. Endless discussions over pull request formatting. Decision syncs. People saying stupid things like "don't let perfect be the enemy of good". Ground-up rewrites because a new language is the fashion. Those were the tolls I paid in order to do the interesting part in the back of the house. I didn't like working with rude people who ran little fifedoms—like I've had to do at some, but (thankfully) not most jobs—but the payoff was worth it.

My new job is to watch the robot work. Then I need to do a ton of reading about what the robot is doing so that I can, at a moment’s notice, seize control back from it and course correct. Calling this "programming" is a farce. I'm closer to those poor souls who have to sit in the driver’s seat of a self-driving car, hands hovering, sweating, waiting for the robot to try and mow down a pedestrian.

In reality, I am a SaaS provider, and the service I provide is responsibility. I am a flat-rate monthly responsibility service that agrees to assume full liability for the output of the robot. I am the human face with a pulse who can be lectured about why something did or didn't work. I will apologize. I will attempt to fix it. I will take the blame.

You Know GDPR Is Good Based on Who Hates It

August 28, 2026 12 min read in opinion

FDR has always been one of my favorite presidents, second maybe to Lincoln. Both were men the establishment assumed were one of them until, to their horror, they governed like they weren't. Both could trash the opposition in one breath and take the moral high ground in the next. One of my favorite Roosevelt lines growing up came from Madison Square Garden, October 1936, standing before a crowd that included plenty of people who wanted him dead:

Never before in all our history have these forces been so united against one candidate as they stand today. They are unanimous in their hate for me—and I welcome their hatred.

What I love is that he doesn't argue with the hate. He doesn't say they're wrong to hate him, or that the hate is unfair. He says the hate is evidence. The process is working. I've always treated it as a metric: if you're doing something hard and nobody hates it, you probably aren't doing it. If the right people hate it and those people happen to be some of the worst people alive, so much the better.

By that standard, the GDPR (Europe's General Data Protection Regulation) is doing beautifully.

It is impossible to go anywhere in a technology space online without hitting a wave of commentary about how stupid GDPR is. It was written by bureaucrats who don't understand the amazing potential of unrestricted technology. These US-based critiques almost always lean on the oldest trick in cyberlibertarianism: we don't have time to regulate, we must simply adapt and ride the wave. Nobody has time for government.

Of all GDPR's consequences, none gets more attention than the cookie banner, which critics present as the inevitable result of government meddling. Blaming GDPR for the cookie banner is like blaming the health inspector for the roaches. The banner is deliberate vandalism, a dark pattern engineered to exhaust you before you can learn anything about the surveillance apparatus humming behind the "OK." Ironically the banner designed to hide the machine has taught the public more about the machine than a thousand podcasts ever will. Even non-technical people stop at "your data is shared with 996 partners."

"For a sports score website?"

So why is the tech commentary community so loud about this? Because they understand what's at stake. If consent must be freely given and easy to refuse, the industry's power shrinks exponentially. People might decide who has their data, how long it's kept, and what it was collected for. You can only imagine how that thought keeps a Meta executive up at night when he's not eating endangered animals, or ignoring calls from his children whose names he has forgotten while on a tacky yacht.

Consider the following example. Did you know Google was doing this every single time you searched on Google? Did your dad? So even in the most maliciously compliant form the regulation does provide value and information.

Remember the hatred is the metric. How did we get here, where US tech companies end up regulated by Brussels? Why isn't the US government regulating US corporations anymore? If the rules are so terrible, why did nobody choose market exit? Has the EU become the world's "privacy cop" or, in the inverse, the biggest player to protect a fundamental human right to privacy?

Ah who doesn't remember where they were on GDPR Day. Since we're all socialists in the EU, we stood up from our government issued desks and gave the required three cheers for regulation, then resumed being on vacation for 6 weeks. Obviously after stopping by my free doctor on my way to the airport. On May 25th, 2018, GDPR took effect to the sound of American commentary, the way fireworks take effect to the sound of dogs.

You can tell American CEOs were aware of the regulation based on the speed by which they copied the language from it. Right before it took effect Brad Smith, the president of Microsoft, tweeted "We believe privacy is a human right." Tim Cook was right behind, telling CNN that "privacy is a fundamental human right". The framing of privacy as a human right is one of the key elements of the EU approach with GDPR. This is in stark contract with the US legal system which views information privacy as more of a market problem. You are all informed individuals in the wide marketplace of data exchanges and are left mostly to your own devices. In theory there should be regulations by the US of things like unfairness, deceptions and other market failures but in practice that doesn't happen.

Europe is no stranger to this fight. The German state of Hesse passed the world's first data protection law in 1970, also known as the year the Beatles broke up, and set a standard we still fail to meet today:

The records, data and results covered by data protection shall be obtained, trans-
mitted and stored in such a way that they cannot be consulted, altered, extracted
or destroyed by an unauthorized person. This shall be ensured by appropriate
staff and technical arrangements.

Link

At the launch of GDPR there were 126 countries with data privacy laws of some sort.

What you see with this sea of legislation is an overwhelming consensus that what GDPR was attempting to do was correct. In fact you see a pretty high level of global convergence of standards. All 126 laws descend from the same commandments the OECD carved in 1980: collect only what you need, say what it's for, keep it safe, let people see and correct it, and don't be a creep about any of this. Fifty years later, the American internet industry is still stuck on commandment one.

So first the often-repeated sentiment that this is a flight of EU fancy is straight up incorrect. Something you could describe as the "European standard" for data privacy quickly became a global standard.

Anu Bradford calls it the Brussels Effect: Europe regulates, the world complies, because despite American bluster, Europe is a market nobody can leave.

In the US financial market, companies who surrendered the EU market would have quickly found themselves with new CEOs as their previous leaders suddenly discovered health problems or a deep love of their families that they had ignored for years. Especially given the increasingly frosty relationship between the US and China, companies that were pushed out of China due to regulation and increased domestic competition cannot lose the EU market. It's also difficult to screen a lot of services for EU customers, especially because the laws follow the personal data of EU residents whenever and wherever the information is transferred outside of the EU. Think of it like trying to sort luggage based on what stickers are on the outside.

The EU is also set up in such a way where enforcement of such a law becomes possible. Every member state has a Data Protection Authority, who are charged with assisting individuals in protecting their rights, advising domestic legislatures on the functioning of existing regulation and finally enforcing the law. The EU is also different from the US in that it is open to exploring precautionary regulatory action. We see this with the EU Artifical Intelligence Act which tried almost immediately to get some controls on the industry right at the beginning. Link So the combination of a robust option for enforcement combined with an increased appetite for regulation in general and a high level of respect for personal privacy made the EU the logical source for this legislation.

Want to know what the teeth look like? In 2011, an Austrian law student named Max Schrems asked Facebook for everything it had on him and got back 1,200 pages, much of it stuff he'd never volunteered. He filed a complaint from a dorm room. Four years later, the Court of Justice of the EU had voided Safe Harbor, the transatlantic data treaty, on the strength of it. A college student complaint killed an international agreement signed by presidents and prime ministers....multiple times.

Japan and the EU

The best proof of GDPR's power is what it did to Japan. On January 23rd, 2019 the EU and Japan reached a deal which allowed for the free flow of personal data between the two economies. It established an overarching privacy law with a core set of individual rights and enforcement by independent supervisory authorities. The process took 2 years, which isn't a surprise because before this process Japan had very weak, swiss-cheese regulations.

In 2014 Graham Greenleaf chose the title "The Illusion of Protection" for his chapter about Japan in an overview of Asian privacy laws. The private sector was basically unregulated, it has "easily manipulated exceptions" to its rules concerning the use and disclose of personal data, its absence of provisions for sentivie information and had no restrictions on data exports. It was nowhere near the level of protections that the EU would expect for information sharing.

You can draw a straight line from a bargaining table in Brussels to new rights for a retiree in Osaka. Japanese data brokers hate it, which again is the point.

Why hasn't the US sought the same arrangement? Because everyone involved knows the application would be denied. Which raises the real question: why can't the country that invented most of this technology produce a rule for it? Why is the US stuck pretending there's no reason to regulate while the rest of the world moves on?

The US and Surveillance Capitalism

There's no better text on what has happened in the US with the data economy than The Age of Surveillance Capitalism by Shoshana Zuboff. It's a good read and I won't ruin it for you.

First, the definition of Surveillance Capitalism from the book.

1. A new economic order that claims human experience as free raw material for hidden commercial practices of extraction, prediction, and sales; 2. A parasitic economic logic in which the production of goods and services is subordinated to a new global architecture of behavioral modification; 3. A rogue mutation of capitalism marked by concentrations of wealth, knowledge, and power unprecedented in human history; 4. The foundational framework of a surveillance economy; 5. As significant a threat to human nature in the twenty-first century as industrial capitalism was to the natural world in the nineteenth and twentieth; 6. The origin of a new instrumentarian power that asserts dominance over society and presents startling challenges to market democracy; 7. A movement that aims to impose a new collective order based on total certainty; 8. An expropriation of critical human rights that is best understood as a coup from above: an overthrow of the people’s sovereignty .

Didn't read that? I don't blame you. In English they found oil and the oil was us. And as us Americans love to do, we immediately discovered a sudden love of freedom in the presence of oil.

Effectively here's what happened. The 9/11 terrorist attacks had a ripple effect in US regulations, effectively derailing the momentum that the domestic US regulatory organizations had about starting to build frameworks around personal data. The focus became on security and not privacy. Quickly public intelligence agencies and the fledgling surveillance capitalist business in Silicon Valley found each other and carved out the concept of "surveillance exceptionalism". If you were spying to keep us safe, then it's not spying it's public service.

As time went on, the corruption of American politics rendered the possibility of regulation less and less feasible at the federal level. Google and Facebook poured tens of millions into lobbying (for non-US readers, lobbying is a nice way of saying bribery that is legal). There also became an "open door" between government and tech, with 197 people moving back and forth from Washington to the Googleplex.

What these companies learned is that by studying our behavioral data during periods of relaxation or play was the most value, allowing them to accurately and reliably push people towards profitable outcomes. US consumers were aware of this to some extent, often repeating phrases like "If it's free, then you are the product". That was true before, but in the new digital economy it is no longer true.

We're not even the product anymore, which is why all these companies no longer give a solitary shit about whether their stuff is good or fun to use. We're the raw material that they mine. They know every single thing about us and we don't know anything about them. They accumulate all the data from us but not for us or for our benefit. Behavioral modification though the accumulation and manipulation of this data is now the wealth engine of the United States.

Now we are in a market failure scenario. No individual company will disarm first, and any regulator can be captured for what these companies spend on catering. Only law with real enforcement teeth changes the math. And the US cannot pass laws with teeth anymore, not because voters don't want them (polling says they do), but because the pipeline is purchased. When someone says America "can't" regulate tech, they mean it the way a hostage "can't" reach the phone. GDPR didn't spread because Europe is heroic. GDPR spread because Brussels is what fills the room when Washington leaves it.

What does surveillance capitalism look like?

Thanks to the amazing https://decryptads.com you can see what it looks like in real time. Let's take The Verge.

This is a relatively straightforward tech commentary website that has a paywall. It should have a very simple supply chain in terms of advertising.

What we see is the opposite. There is a giant network of companies trading, selling and bidding on the information from this website. This isn't the fault of The Verge, this is just how the machine operates. Your data is like fish at a fish market. Everyone gets to inspect the merchandise and decide whether they want to buy without you being involved. Ironically we only get this information because of the ads.txt and sellers.json which exist to combat advertising fraud.

Even a well-run website operated by technologically literate people geared towards tech enthusiasts behind a paywall is not immune to this system. What you see here is that there is no fucking escape. If you want to have a large web presence and have bills to pay, you need to participate. And The Verge is doing it right! 59 declared partners and none of them are actively preparing for war with the US.

In the US though this is not a new problem.

Double Movement

The economic historian Karl Polanyi came up with the concept of "Double Movement". You can read it here.

Think of it like this. Imagine you strip American Capitalism down to a tug of war. In the first round, the businesses go first. This is the "let the invisible hand decide" part. Everything becomes a product you can buy and sell including land, work, even money itself. Think of this like removing all the referees from a game and letting players do whatever they want.

Then there is round 2, which is "wait this is hurting people". People demand protections like minimum wage laws, workers rights and trade regulations because the "free market" never seems to regulate itself but is causing real harm. The referees come back, but now with a rulebook written by the players who got hurt.

His argument is that a totally free market is a fantasy. It's a fantasy because markets need government support to operate. Even people who pretend they despise government interference rely on them. Copyright and trademark doesn't matter when you need to train an LLM, but it matters a whole lot when I start marketing my iWatch smart watch.

That rhythm of abuse, then correction governed American capitalism for a century. For surveillance capitalism, round two never comes. There is no functional federal counterweight as I write this in 2026. Some states are trying, and good for them, but regulating the internet state by state isn't progress it's just stopping the bleeding. Meanwhile the brokers buying and selling your life exist entirely outside your scrutiny, have no fear of comprehensive reform, and can swing a statehouse election for what they spend on a quarter's catered lunches. Entities this powerful cannot coexist with a functional democracy — a fact they seem to have considered, given the rise of the "Nerd Reich". https://www.npr.org/2026/08/10/nx-s1-5925350/the-nerd-reich-tracks-the-unmasking-of-silicon-valleys-true-politics

So the order of operations is simple. Washington can't act, so Brussels does. Brussels acts, and the world follows, because the market is too big to leave. That is the whole story of GDPR: not European ambition, but American absence. A sign of the declining empire if you will.

Roosevelt gave that speech at Madison Square Garden on the last night of October 1936, and a week later he won forty-six states. The people who hated him got Maine, Vermont, and the next ninety years of being wrong about everything they hated. That's the thing about being hated by the right people which is it is a currency and a valuable one.

Privacy regulation will get there too. Not because the industry repents because industries don't, but because this is how every one of these stories ends: seatbelts, smoking sections, lead paint. Normal, then scandalous, then unthinkable. Someday someone will ask what an ad network for children was and refuse to believe the answer. And the executives who fought this will be on a boat somewhere, explaining that they were for it all along.

Accept all.

In Purgatory Everyone Likes Crepes: My Time at a Greek All-Inclusive

August 21, 2026 15 min read in humor

I can’t remember the last time I was actually hot. It’s a thought that keeps looping through my head as I stand next to a children’s play structure in a Greek all-inclusive waterpark. The sun is beating down on an international community of mostly parents drinking watered-down cocktails out of plastic cups while their children scream and run in circles. Surrounding this little paddock of plastic is a lazy river, filled with mothers and fathers staring blankly into the middle distance as they slowly orbit the water toys. It’s all concrete and aggressively bright primary colors, suffused with the smell of gyros being cooked by a Filipino staff behind me.

Living in Denmark, it’s rare for the sun to shine directly on you. Typically, everything outside looks like it has a blue filter placed over the lens—like how TV shows put a yellow filter over the camera to establish that you are now in Mexico. Sometimes in Copenhagen you might get hot, but then it will immediately start to hail on you, and you suddenly have a different, worse problem. Greece, by contrast, feels like the sun is noticeably closer to the earth. It feels personal, like a heat lamp pressed against your skull.

My daughter has been playing peek-a-boo with a cute baby on a bench next to this water playset for long enough that I feel a moral obligation to go over and say hi to the mom. She’s a nice woman, watching four kids at a waterpark alone with a casual calm I actively envy. I’m a little stressed keeping eyes on just one, but she has four kids ranging from around ten to two, running around and disappearing beneath the waves, only to emerge just before I would panic. She has a small sunburn on the small of her back that I assume she either can’t reach or forgot—a small, raw triangle of red.

We chat briefly. She asks where I live; I ask the same. I’m surprised when she answers Russia, mostly because I don’t get much exposure to Russians in Denmark. I clearly make a face without meaning to, because she quickly adds, "Their father is off fighting in the war."

Ironically, I think she added this to head off me judging her as a single mother. Instead, revealing the father is a Russian soldier has made the part of my brain that regulates politeness completely short-circuit. What is the social etiquette for the wife and children of a military power you don’t support? I’ve never even considered this question, standing here in my swim trunks, my too-pale stomach exposed to a hostile world. I don’t want her children to be orphans, obviously. Neither she nor they did anything to start this conflict. But I’m worried that expressing vague positivity will end like my conversation with an Israeli in Copenhagen who took my neutral statement, "I hope the conflict ends soon," and responded with a wink and, "It will, with Trump at the helm."

"I hope he comes home soon," I say, keeping my face perfectly neutral.

She nods and resumes throwing a ball around with her children. She seems completely unaffected by thinking about her husband, which I think is actually the strange magic of the all-inclusive resort. It is a geopolitical anesthesia. The world is burning down, but here, the ice cream machine is still running.

On most topics, I don’t consider myself a snob. I enjoy a Michelin-star restaurant, and I enjoy a nice slice of 7-Eleven pizza at 2 AM. I'll finish a thick, sad Classic of Literature and then immediately pick up Space Marines vs. The Metal Skeleton Wizard on the Moon. But on travel, I’ve always looked down on the all-inclusive crowd. Sitting one rung above a Carnival Cruise on the ladder of human decay, the all-inclusive always seemed like the most cynical way to say you had legally stepped foot in another country.

My first real exposure to this was in Puerto Rico. I would leave my bespoke 20-room boutique hotel and scoff at the barbed-wire fence surrounding the American mega-resort down the road. "Why would you fly to the Caribbean to eat at a steakhouse?" I’d say, dripping with scorn to my wife, who would nod as we walked through dark neighborhoods on our way to eat dinner in someone’s actual living room that Google Maps called a restaurant. As I traveled the world, I saw these giant compounds everywhere. I swore I’d never become one of those placid, sunburned people wandering around a sculpted grass lawn, slightly buzzed at 11 AM.

So imagine my surprise when I found myself finishing my second tequila sunrise before lunch, staring at a too-blue pool. The resort’s "activity crew" was out on Crepe Island—so named because it had a full crepe station on an island in the middle of the pool.

"Really, it's a crepe peninsula," I say to nobody. Nobody cares. They are aggressively not paying attention to anyone else. This is not a community.

A line of unlicensed, Disney-and-Nintendo-adjacent costumes are paraded out, with "Merio" closest to me. Kids kept trying to give Merio a high-five. I don't know if it was the suit or if the guy inside was just sick of his job, but he kept waiting too long to high-five the kids back, inevitably wacking them in the face or the back of the head as they moved on. Over and over, an excited child would run up, followed by a bizarre, lagging delay before the inflated hand would strike. "At some point, that's just hitting kids," I said to the ether. The woman next to me put her AirPods back in.

My wife had booked this trip as a reward to our daughter. Having been a remarkably good sport as we dragged her to such famously child-friendly vacations like "A Sheer Cliff Overlooking Icebergs in Greenland" and "Endless Climbing Up and Down Stairs in Florence," we wanted to give her a chance to just be a kid. When we asked her what her favorite vacation was, she responded with zero hesitation.

"Lalandia," she’d say, before resuming coloring in her coloring books with so much force you could see the head of the marker get pushed back inside the plastic casing. Since Lalandia is effectively a massive, indoor all-inclusive in Denmark, we figured the same concept in Greece would at least be a warm alternative.

We boarded a chartered flight direct to Greece, my first time on a flight where everything was delivered in Danish. I felt like a spy hiding among the locals, fearful they'd discover the American among them. Safe from prying eyes, I was pleased to watch the Danes engage in behaviors they usually made fun of us for. Loud conversations were had, crunchy snacks were aggressively unwrapped, and general uncouth behavior typically shunned was suddenly fair game at 30,000 feet.

Upon landing, the tour coordinator told my wife and me that we had to "run" to make our bus. My wife took off for the toilet; I grabbed the kid and the big suitcase and sprinted for the van. We then sat in the van waiting to take off for over an hour. What I hadn't realized in my heroic rush to the van was that I had left a small carry-on sitting in the middle of the airport. It was filled with snacks, coloring books, and also my wife’s wallet and passport.

The first night at our all-inclusive was understandably tense. Variations on the word "idiot" were thrown around. The next day, we got the good news that the bag was safe and sound at the airport police station on the island of Kos. So while my family went off to the pool, I went to the lobby and called a cab.

My driver arrived 30 minutes later and immediately asked, "Is it ok if you sit up front so we also take these beautiful people downtown?" The Scottish woman in the back blushed at this. I, lacking the social architecture to recognize warmth, simply thought: Wow, bold move saying that in front of her husband.

I got into the passenger seat. This, as I realized later, totally broke down the driver-passenger social construct. After dropping the charming Scottish couple off in the downtown area, my driver started telling me literally everything about his life. For the next forty minutes, I learned how his brother had been a firefighter, then one day went on vacation to Vietnam, fell in love with a Vietnamese woman, and opened the only authentic Greek gyro restaurant in Hanoi. He had a booming, theatrical laugh and was poured into a skin-tight polo shirt.

"Why are you going to pick up this suitcase your wife lost? DO WE EVEN NEED TO ASK?? HAHAHA! WOMEN!" Then he’d get deadly serious and remind me, "Children, they're the most important thing." I had spoken maybe six words at this point in the trip.

I forced a laugh, silently calculating the statistical probability of being murdered and buried in a Greek olive grove. It would be a beautiful place to die, though I worried they’d use a photo of me from this trip—pale, squinting in a baseball cap—for the Netflix documentary thumbnail. Some true crime podcast would inevitably conclude I had it coming.

At one point, he pulled the cab over to show me a flat patch of grass. "This is where I saved the island from a wildfire," he said proudly. I couldn't see any sign of fire; it just looked like a normal field with the grass crushed.

I didn't say anything, as up to this point, the conversation had been delightfully one-sided. I had mostly sat sweating through my shirt, nodding. He told me how someone had thrown a cigarette out of their car, and it had started to smolder. He had stopped and used his legally required fire extinguisher to put it out.

"It's good that I had this," he said, patting the dash. "This is why a robot can never replace taxi drivers." He then stared at me as if I were an assassin sent from Silicon Valley.

"You betcha," I said, sweating. I wanted to assure him I wasn't there to replace him with an app. My amoral peers back home were already working on it, but I personally wouldn't kill his livelihood. Just people who looked and sounded exactly like me.

I kept poking around silently instead, but then he caught me looking at the giant bulldozer sitting twenty meters away, its tracks leading directly to the field. I'm a genius like that.

"Yeah, alright, also the bulldozer helped a bit to put the fire out. But mostly, it was my fire extinguisher." The bulldozer was merely an accessory after the fact.

He then changed the subject by telling me that Tom Hanks had sat in his taxi "right where you are sitting." "Did you know Tom Hanks loves Greece and we love him? He is such a friend of Greece that we gave him a passport." Is Tom Hanks Greek? I thought, staring out the window. Why are there so many interior decoration stores? Is everyone on this island redoing their bathroom?

"Tom is so nice, we love him." My brain returned absolutely no information about Tom Hanks, which frankly is typical of my mind. He clearly wanted me to respond with a Tom Hanks factoid, but all I could think to say was, "Yeah, his son was really good in Fargo."

"HIS SON? Who is talking about his son?" He took the conversation back over, explaining how Tom Hanks’s wife had sat in the back of the cab while Tom sat in the front. Weird choice, Tom, I thought, looking for a sign we were wrapping this trip up.

When we arrived at the small airport, he parked randomly outside the main building, basically just the corner of the road that led from the airport back to town. "Wait, are you coming in with me?" I asked, confused.

He nodded, saying it would "take way too long" if I went in by myself. Suddenly we were concerned about efficiency. He marched us in, knowing everyone who worked there. I found myself in a small police office that definitely didn't understand the separation of church and state. There was a massive crucifix on the wall, and each police officer's desk was covered with a Jesus mousepad and a large picture of the Virgin Mary. My cab driver and the police officer negotiated the release of the suitcase in front of me without bothering to involve me. At some point, I was told I could take the suitcase and get out of here. Nobody had asked me any questions except to quiz me about the contents of the bag.

On the way back, we stopped for tea because, truly, what is a taxi meter at this point? We'd been in the car together for like two hours. He told me how his mother-in-law had almost died from a heart attack. "It's good that she could get to Athens in time," he said. He then told me how during the off-season he harvests olives from his family farm and has them pressed down the street. "It's heaven working the fields with your family."

I imagined my siblings and I being asked to harvest olives in the beating sun and immediately envisioned four body bags in the shade of an olive tree.

I nodded, staring out the window at a fire extinguisher store across the street, where what looked like a 12-year-old boy rolled a cigarette seemingly one-handed and lit it while sitting on top of a barrel. The long grass around the fire extinguisher store seemed primed for a fire. What happens if a wildfire hits a fire extinguisher store? I pondered, while my driver went on about the beauty of the Mediterranean or something I wasn't paying attention. Then we packed it up and went back to the hotel.

"Thanks for a nice morning, Mark," he said fondly to me as I got out. My name is Mat. But I guess if you've saved the island from a wildfire, you can call me whatever you want.

I got back just after Mini Disco. This was a one-hour dancing marathon for the kids in the semi-enclosed theater. With two fully manned drink stations on either side of the stage, it was an opportunity for children to dance while their parents drank like an asteroid was imminent. The songs stayed the same every night, but on this first night, we were going to learn a valuable lesson: Always buy the t-shirt.

Apparently, while I was learning the ins and outs of the Greek cab business, a nice woman had asked my wife and daughter if they wanted to buy a t-shirt for the Mini Disco. My wife, smelling a tourist scam, had hard-declined. What she didn't realize was that the climax of the Mini Disco was a formal t-shirt presentation ceremony. Each child's name was called, they were presented with a t-shirt, and then they were allowed to hug Leo the Lion.

My daughter didn't understand that she didn't have a t-shirt. So, when the name "Eleanor"—which is not her name—was called, my daughter hopped up and snagged the shirt. This meant my wife had to rush onto the stage, rip the t-shirt out of our daughter's hands, and hand it to the sheepish little girl whose shirt it actually was, all while holding a cocktail. It was a masterclass in parenting under the influence.

The next day, we were on the hunt to put in an order for the t-shirt, having made the most serious promises parents can make to a child. After we chased down the red-haired French woman running the kids' merch table, we went to the main pool. She didn't seem surprised to see us come crawling back. I got the sense the first day she was doing the sales pitch, then after that she just waited for us to come to her.

There is a weird etiquette to pools and British people. They will rush out the second you are legally allowed to put a towel down on a chair, meaning from 8 AM to 1 PM, there isn't a single open seat. The British treat pool chairs the way trench soldiers treated no-man's-land: as hotly contested territory worth dying over. But if you are lazy like my family is, you just wait. They start to leave the pool around 1 PM, and you can get a great seat with zero work. The next day, you repeat the entire cycle. That evening, kiddo got her shirt. It was a proud moment; she clutched her blaze-pink t-shirt to her chest and teared up with pride.

Soon, the days started to blend together. At some point, we all got an emergency text message telling us about wildfires on a neighboring island. I assumed the resort staff would need to calm down the crowds, as you could pretty clearly see the smoke from the fire. The sky was turning a dark, orange-brown in the distance, and the air smelled like burning rubber.

Nobody cared at all. A thousand people looked at a Greek text message they couldn't read, put their phones down, and picked up their beach reads. The apocalypse was met with a shrug. As an allegory for global warming, there is something particularly bleak about people dancing in a pool to 90s boy band hits as a thick plume of wildfire smoke shoots up into the sky. Thankfully, here nobody pretends to care about anything, so I was able to slip back into a soothing apathy. In a world that was constantly asking me to pay attention to some fresh horror, this was a place that asked nothing of you.

Toward the end of the trip, we decided to head down to the small town and were dropped off at Dolphin Square. We started walking around, looking at Google Maps to see what we should look at. There was a Roman House, which is basically an empty lot full of pieces of old Roman architecture and a sign saying I needed to pay 10 euros a person. Since nobody was there to collect this 10 euros, the sign felt more aspirational than practical. Someone should pay this Greek island 10 euros, but not today, I guess. But I looked at the old rocks.

My daughter was unimpressed. She looked at the ruins, looked at me, and asked if she could get ice cream. "Denmark has a lot of old stuff," she said, which, in her defense, is true. We walked around a bit more, buying little tourist trinkets. I'll never understand who buys the t-shirts that fill the small streets of these towns. Who is the demographic for a shirt that says "I heart my boyfriend"? And more importantly, where is he? Does he buy the t-shirt for you before you go on a solo trip without him?

But we went to a small restaurant and had a proper meal for the first time in days. Despite the heroic efforts of the staff at the resort, it was actually hard to eat there meal after meal. The culinary philosophy seemed to be "cook the will to live out of it." Everything was either fried and dried out beyond belief, a hard puff pastry, or so bland you couldn't really tell what was going on. The lamb sorta tasted like the chicken, which tasted a lot like the pork, which tasted like a cry for help. At one point, we went to an Asian-themed restaurant on the resort, and I was served "duck with Chinese pancakes." The duck was well-cooked, but the pancake was an Old El Paso flour tortilla. I couldn't help but feel like the duck died for no reason.

At one of these meals, where my daughter pounded another plate of french fries and pizza, a British woman at the table next to us told me how her entire family looked forward to this trip. She started to tear up as she explained that she and her husband had worked extra hours this year to make it so their kids could come with them. "It just makes me burst with pride to think about," she said. As she spoke, one of her kids sat beside her, wearing noise-cancelling headphones, staring at an iPad, and methodically eating an entire pepperoni pizza without making eye contact with another human soul.

By the end of our week, I was eating meals of plain bread with watermelon and coffee. My daughter was consuming what looked like a kilo of Nutella, and my wife and I settled down for another cycle. We only ended up going to the actual beach, which was beautiful and maybe 200 meters from our hotel room, once. The ocean was beautiful, with rolling hills in the background, perfect warm water that was the saltiest water I've ever been in. It was like swimming in a giant, warm tear. Everyone but me hated it because it wasn't as nice as the pools.

"Ugh, there are ROCKS!" my daughter shouted, upset at the audacity of the ocean for existing. Honestly, the rocks hurt a fucking ton, but I was too self-righteous to admit it. "Let's enjoy nature, everybody!" I yelled, bleeding from the feet.

I thought by the end of my time with these people that I would come to think less of them. I expected to leave despising their complacency. Instead, I found myself weirdly protective of these folks. They are just trying to make some childhood memories with their kids, trying to manufacture something, anything that looks like a normal childhood in a world burning down.

I won't lie, the all-inclusive life isn't the life for me. It is a place out of time, where the enjoyment of it requires a total suspension of your relationship to the outside world. But I do now understand the appeal of not being challenged. In a time when everything is being questioned and every tradition and norm is falling apart, this is a callback to frankly an easier time to be alive. And we ended up getting the t-shirt, which, when you break it down, is really the most important part.

OTel Isn't Going Well (And I Made A Spreadsheet About It)

August 12, 2026 14 min read in opentelemetry

For years now one of the most reliable complaints I hear when I try to drag a team off their vendor specific SDK and onto OpenTelemetry is some variation of: "why does it seem like this isn't done yet?"

Vendor SDKs for observability are, to put it charitably, idiot-proof. You install the thing, dashboards just load data, someone else worries about how all those pieces fit together, and you get on with your life. OpenTelemetry, by contrast, greets you at the door with a lot of "experimental" stamps and roughly six different ways to accomplish any given task.

In OpenTelemetry's defense this was never what they were going for as a project. I've always respect that they stuck to their guns by attempting to build a truly vendor agnostic system that really doesn't care what you do with the data. I have never gotten a sense of a vendor being strongly preferred with OTel, which is quite the feat considering how lucrative and contentious the observability ecosystem was. Also considering that the maintainers of this project are largely employed by exclusively those companies.

As the years wore on, I started to get nervous. Conversations in the semantic-conventions repo drag on and on and on. Different languages had dramatically different stories. Golang and Dotnet were first class citizens, but other languages lagged years behind the others.

I started asking a lot of probing questions before recommending OpenTelemetry to smaller teams who didn't have the time, budget, or emotional bandwidth for it. Auto-instrumentation was genuinely magical, but the cliff between "auto-instrument works" and "now I have to manually instrument something" was steep enough that you owed people a warning before you pushed them off it.

This narrative has been going on for awhile in the observability space, a vague sense of "something is wrong in Otel-land". But let's try to generate some actual data here. Is there an actual problem, or is this something where the perception by the community of slow progress is imaginary? Is the problem not enough maintainers, too big of a scope, or something in-between?

My guess when I started was "oh this is your classic open-source bit off more than they can chew". Not enough maintainers, not enough budget. Now there is some of that, but there's also something else going on.

The actual problem happening inside of OpenTelemetry is a three way crash. You have a binary stability gate which, when combined with a very small bench of actual maintainers means there is understandable worry about marking a feature not experimental then add on just a massive scope of languages and frameworks they are attempting to cover. This creates a perfect storm where there is an incentive to argue about potential problems a feature might create since once it is locked in and shipped as stable you can never change them.

How does OpenTelemetry Work

So OpenTelemetry currently is attempting to support a dizzying number of languages and frameworks.

OpenTelemetry is a giant project. It spans dozens of languages, hundreds of libraries, and countless backends. To keep things sane, the project splits work into two buckets:

Core → Maintained directly by the OTel project. Small, stable, vendor-neutral, and tightly reviewed. This is the "spec-defining" surface.
Contrib → Community- and vendor-contributed. Broader, faster-moving, and covers the long tail of integrations.

There exists the otel-collector, the thing that runs along the thing so that you can ship logs metrics and traces. That copies the same rough pattern. But for the languages when we're talking about core vs contrib this is what we're talking about.

`opentelemetry-python` (core)	The API, SDK, OTLP exporter, context propagation, resource detection primitives
`opentelemetry-python-contrib`	Instrumentation libraries for Flask, Django, requests, psycopg2, Redis, Kafka, boto3, etc.

Stuff that breaks goes in contrib, stuff that doesn't break goes into core.

Now the reason this causes a conflict. contrib is massive overkill for most projects. You don't want 300 exporters to add the one you typically need. On the language side, this isn't that big of a problem. pip install opentelemetry-instrumentation-flask gives you the stuff you need for flask. However on the collector side you end up having to do the OpenTelemetry Collector Builder to make your own collector (or just kinda ride the wave and hope it works out). While cool that this exists, it's a lot of scope to ask a team to take on.

Process of adding a new feature

So I believe I have captured the workflow of adding a new feature to OTel. You can check my homework here:

OpenTelemetry Enhancement Proposal (OTEP) (https://github.com/open-telemetry/opentelemetry-specification/tree/main/oteps/)
Once the OTEP is accepted, the text goes into the Specification directory in the same repo.
After that it seems to go to Semantic conventions. This seems to be where we get down to the specific details and where most of the long discussions seem to live. At this point we're talking about more or less a permanent commitment to this design and where the lock-in process becomes very hard to change.
Each of the SDKs implements the API surface that is defined in the specification. Now some of the SDKs have done 2.0 breaking changes, so it does seem like the earlier "please no 2.0 at all costs" sentiment has been abandoned (which I think is smart and good).
Contrib / instrumentation. This is slightly more mushy. Looks like they should track latest API/SDK but each contrib package may version independently so its more flexible as a design.
Collector + OTLP. The data has to actually go somewhere. OTLP (wire protocol) has its own stability lifecycle and specification (here). Collector components have their own stability in their READMEs and as far as I can tell that's kinda all over the place.

Things I'm not really clear on

It's unclear how long the OTEP -> Specification process takes. I've looked through the Git history but there doesn't seem to be any predictable number or cycle.
I don't fully understand what is the relationship between all these stability commitments. Does Collector + OTLP group work in lockstep? Can a language "fall out of scope" if you lag too far behind?

Attempting to test it

So because OpenTelemetry is a CNCF project, I figured it made the most sense to compare them to other CNCF projects. My basis for comparison is Envoy and Prometheus. I have used a hacky Python script I've used before for measuring the "health" of open-source projects, which is probably not the best. However I'll include a link to the raw data without the charts so folks can review it and (more than likely) find a problem in what I generated.

So we look at 24 months of activity for Envoy and what we see is a pretty healthy project. There's good distribution of authors, mergers, issue closers. phlax is obviously pretty important to the project but in general there's a good bench of people to step in if needed. I've attempted to filter out all the known bot traffic.

Let's compare that to one of the OpenTelemetry languages. The ones I have the most professional experience with are Golang and Python, but I hear from a lot of folks in the community that the Ruby and PHP ones struggle a lot. This is the PHP one for the same period.

So we see pretty clearly that there's way too much concentrated on 2 people. This is not a healthy open-source project and they clearly don't have enough people to cover the kind of scope OTel needs to cover. Same story with Ruby.

In comparison the "strongest" OpenTelemetry SDKs in my opinion, Golang and Dotnet (although Python is also no slouch) look more healthy.

Golang

So the first issue is maybe the least surprising. There's too much concentration among too few maintainers. Your authors shouldn't also be your mergers and your issue closers. Ideally these tasks should be distributed out more evenly.

For what its worth I think the maintainers have done a good job of attempting to keep their discussions public. It was very easy for me to find the public meeting notes of the different groups of maintainers, read through them and see what was going on. I don't get the sense that these maintainers are trying to stop people from getting involved as much as the expectations of stability have, more or less, frozen the project in place.

The issue is more a classic case of "someone has to pay the maintainers". The project is too complex for someone to realistically do this as a hobby. I think any project signing on for such long stability contracts cannot turn to the community of hobbyists expecting assistance. I can't join calls and do the things I would be expected to do for a project of this size and importance for free. But it also means that the people doing this critical work have expectations placed on them by their parent organizations.

Repository	24mo Merged PRs	Distinct Mergers	Top-1 Merger %	Top Merger Role
`opentelemetry-cpp`	544	4	86.1%	Single human (`marcalff`)
`opentelemetry-kotlin`	281	2	79.7%	Single human (`fractalwrench`)
`opentelemetry-browser`	102	4	79.5%	Single human
`opentelemetry-ruby`	213	5	78.7%	Single human
`opentelemetry-js`	829	14	64.9%	Highly concentrated
`opentelemetry-python`	486	4	61.4%	Single human (`xrmx`)
`opentelemetry-php`	181	2	53.0%	Two mergers total
`semantic-conventions`	911	9	49.7%	Single human (`lmolkova`)
`opentelemetry-go`	686	5	36.9%	Distributed bench
`opentelemetry-dotnet`	657	6	31.5%	Distributed bench
`prometheus`	1,849	31	14.4%	Broad bench
`envoy`	5,432	28	35.8%	Broad bench

So these SDKs have too few maintainers. But that doesn't fully explain why it seems to take so long for new features to get through the stack. My guess for that was that somewhere in the process between submission of the new idea and the formalization of the idea was a long discussion that took a million years.

Conventions about Semantics

So with this level of surface area across different frameworks and languages, it makes sense to concentrate the conversation about conventions in one place. That lives here: https://github.com/open-telemetry/semantic-conventions

If vendor debate is causing the slowdown, we should (in theory) see this slowdown in PRs here. Then you should see the slowdown basically propagate out. Spoiler alert, I was wrong about this. Big thanks to the OpenTelemetry people for having good conventions on labeling their PRs which made this much easier.

So if semconv is the slowdown, let's look at the slowest PRs there.

PR	days	comments	reviews	labels	topic
#2083	277.5	17	115	area:gen-ai	MCP semantic conventions
#2617	258.6	29	13	area:gcp	GCE instance labels
#1698	187.9	3	7	area:azure, breaking	rename `azure_` → `azure.`
#2619	174.6	24	8	area:gcp	GCE instance group manager
#3118	147.1	19	8	area:graphql, breaking	GraphQL Recommended vs Opt-In
#1741	141.0	4	23	changelog.opentelemetry.io	Mainframes
#1784	127.3	7	48	area:k8s	k8s.container.status metrics
#2287	118.5	12	95	area:rpc	ONC/Sun RPC + NFS metrics
#2179	117.0	7	114	area:gen-ai, breaking	Gen-AI chat history attributes

Yeah some of them are pretty slow, but there are some complex topics being discussed. However interestingly this slowdown doesn't really trickle into the SDK/API space, suggesting that OpenTelemetry is going a good job of keeping these conversations siloed off.

If we look at Python we see that their slowest PRs aren't semconv related.

PR	days	comments	reviews	labels	topic
#4646	361.1	5	19	—	OpAMP integration sketch
#4576	314.0	11	27	Stale	OTLP HTTP max_export_batch_size
#4609	253.7	2	9	—	env carrier
#4709	172.1	8	40	—	http exporter error handling
#4333	164.9	6	4	—	GRPC exporter backoff config
#4654	161.7	7	14	log-breaking-changes	deprecate events API/SDK
#4863	155.0	5	36	—	add/remove metric readers at runtime
#4647	152.9	4	9	Approve Public API check, log-breaking-changes	rename Log → LogRecord
#4854	150.5	7	9	hold	W3C traceparent random-trace-id
#4676	126.4	20	30	Approve Public API check, log-breaking-changes	logs SDK refactor

In reality the slowdown for these are the extra required check imposed by the Approve Public API check which requires another maintainer. But that seems appropriate and takes us back to the initial problem of "not enough maintainers".

Potential Solutions

So after looking at all of this, the pattern becomes clear. A new feature takes a very long time to make it to the end user in OpenTelemetry because they take stability very seriously, combined with a relatively limited bench of talent to pull from. Once things make it through the entire stack, implementing the API and getting that API change through to the end user falls on an overworked maintainer pool. So what do we do?

I think one idea worth exploring is adding some sort of time-bound beta tier. Basically between the "Experimental" and the "Stable" in the following diagram. The problem is that for end users, due to the extra steps to use Experimental features, they might as well not exist. 99% of us have no idea when an experimental feature is added and we would never engage with it. But if I knew the feature would stick around for at least 12 months without a removal and was more accessible to me as an end user, it could actually help the project get more actionable feedback.

Basically a feature would go Experimental (pretty low usage) -> Beta (more exposed to the end user than Experimental) -> 12 months -> Removal or Stable.

Now confusingly Beta exists for Otel but is used for SDKs, not for components. Like Rust is a Beta but it seems like Profiles cannot be a Beta. Honestly it's nearly impossible for me to figure out like what labels should apply to what things. I suspect nobody really knows. Here's the explanation of Beta that I think only applies to SDKs.

Development

Not all pieces of the component are in place yet, and it might not be available for users yet. Bugs and performance issues are expected to be reported. User feedback around the UX of the component is desired, such as for configuration options, component observability, technical implementation details, and planned use-cases for the component. Configuration options might break often depending on how things evolve. The component SHOULD NOT be used in production. The component MAY be removed without prior notice.

Alpha

This is the default level: any components with no explicit maturity level should be assumed to be "Alpha". The component is ready to be used for limited non-critical production workloads, and the authors of this component welcome user feedback. Bugs and performance problems are encouraged to be reported, but component owners might not work on them immediately. The component's interface and configuration options might often change without backward compatibility guarantees. Components at this stage might be dropped at any time without notice.

Beta

Same as Alpha, but the interfaces (API, configuration, generated telemetry) are treated as stable whenever possible. While there might be breaking changes between releases, component owners should try to minimize them. A component at this stage is expected to have had exposure to non-critical production workloads already during its Alpha phase, making it suitable for broader usage.

Release Candidate

The component is feature-complete and ready for broader usage. The component is ready to be declared stable, it might just need to be tested in more production environments before that can happen. Bugs and performance problems are expected to be reported, and there's an expectation that the component owners will work on them. Breaking changes, including configuration options and the component's output, are only allowed under special circumstances. Whenever possible, users should be given prior notice of the breaking changes.
Stable

The component is ready for general availability. Bugs and performance problems should be reported, and there's an expectation that the component owners will work on them. Breaking changes, including configuration options and the component's output, are only allowed under special circumstances. Whenever possible, users should be given prior notice of the breaking changes.

Deprecated

Development of this component is halted. No new versions are planned, and the component might be removed from its included distributions. Note that new issues will likely not be worked on except for critical security issues. Components that are included in distributions are expected to exist for at least two minor releases or six months, whichever happens later. They also MUST communicate in which version they will be removed, either in terms of a concrete version number or the date of a release, like: "the first release after 2023-08-01".

Unmaintained

A component identified as unmaintained does not have an active code owner. Such components may have never been assigned a code owner, or a previously active code owner has not responded to requests for feedback within 6 weeks of being contacted. Issues and pull requests for unmaintained components SHOULD be labeled as such. After 6 months of being unmaintained, these components MAY be deprecated. Unmaintained components are actively seeking contributors to become code owners.

In addition it is, respectfully, misleading to imply that Go and Ruby are being maintained at the same standard. This isn't a shot at the Ruby folks — they are doing heroic work with what they have. But pretending parity exists when it doesn't just creates confusion and quiet resentment when a user shows up expecting one experience and gets another. Being honest about maintenance tiers would let people make informed choices and might attract more help to the other tiers by naming the problem out loud.

Finally I would try to surface these problems more openly for OpenTelemetry from the perspective of "we need more maintainers". I feel like the people doing this work probably knew there was a problem, but it seems like the community at large has no idea that there is a need for frankly more engaged ideally independent maintainers and contributors.

OpenTelemetry is a great project that is doing great work. It's doing, frankly, heroic work at this scale with this few people. But I think in order to actually replace the vendor specific SDKs we need to start getting a bit more pragmatic about what is realistic to do in terms of stability contracts and number of languages. I don't think breaking changes are as devastating to the community as these promises imply as long as they are communicated well and I think with this thin of a bench of maintainers, something has to give.

Anyway feel free to check my data for accuracy and let me know if you find problems!

data

data.zip

343 KB

I added a blogroll

July 20, 2026 1 min read

I realized that it might be nice if you happen to stumble on this website if I had a way to recommend other websites you might enjoy. As it turns out this is a "blogroll", a concept I have never heard of before today but whatever. It's still a good idea.

I tried to add some CSS to make it easier to follow and search, but let me know if you think I missed a great site that people should check out. I'm also always on the hunt for more good stuff to read. You can find my email and social on the About page above.

Also if you want to just take this Ghost theme and use it yourself feel free: https://gitlab.com/matdevdug/minimal-ghost-theme

Anyway here is my new blogroll: https://matduggan.com/blogroll/

Art Doesn't Scale

July 17, 2026 8 min read in rant

Listening to two acquaintances argue about if AI art is "art" is not a new experience for me, the now eternal debate between Doomer and Boomer. In this particular context, I'm listening to it as I wait for my surprisingly expensive iced latte in downtown Copenhagen. This coffee shop is one of my favorites, a small hole in the wall where the shop starts on the second floor. Everything is hand painted in a way I associate with the hippie roots of Copenhagen. They have an older golden retriever lying on the floor in the heat with a sign on his crate telling people not to bother him if he's taking a break. "Sometimes he needs his alone time". Me fucking too I think as this conversation drags on.

There is a certain irony to the people having this conversation. One is a well-off tech worker, telling the much less well-off artist that the thing he has made through a prompt and the thing she practiced for years to make are functionally the same thing which is a classic AI Boomer argument. "I'm not saying I would sell mine, but isn't the point of art the emotion it brings out when you see it?" The artist seems somewhat baffled by the question, which I understand because she doesn't see the world as a machine that inputs ideas and outputs profit. I use some LLM tools while programming but I would argue I am still more Doomer than Boomer.

As someone who straddles these two worlds, both working for large corporations and interfacing with these borderline sociopaths in business casual pants and expensive watches and also as someone who finds the joy in their life through the celebration of the artistic output of others, I understand them both. One sees it as a capacity problem, a barrier broken down so that now he can do everything and no choices have cost. The other sees it as a baffling attack on her passion. You could always learn to draw, nothing was ever stopping you is her refrain. Which is true but doesn't resonate with him, in part because I suspect he believes he could do anything.

But I hear this argument enough that I thought we should talk about it, if for nothing else than to give me something to think about as I stare at the dog crate and hope the dog comes out to say hi.

The Smell of Ink and Nerds

So I love books of all sorts, including a lot of very trashy books. Nothing I enjoy more than a Space Marine ripping the head off an alien. One medium that I really like as an adult with limited free time are comic books. I love everything about them, the feel of the too slippery covers and the rub-off on your fingers of old comics. I love how weird they are, how stories that don't work are dropped on the floor and never mentioned again. It's like watching someone build a railroad as the train rides down the tracks.

My routine for years was to go pick up my pull list and then wander around for a bit while they assembled it. This is how I found great series like Sandman and countless others that ended up being some of my favorite reads. One of these pull list discoveries I made while standing around surrounded by the delightful acidic smell of a comic book shop was The Black Monday Murders. You can find it here.

The story of The Black Monday Murders is genius and I won't ruin it. You can read the first issue for free at the link above. My friends and I still text each other "All Hail God Mammon" when the stock market goes way up or down. The basic concept was "what if money and magic were related and banks were basically worshiping the god of greed". I consumed this series and fell in love with it.

I mean come on what's not to love about that.

So the first issue comes out in 2016, we get a....very slow drip of new releases through 2018. 2018 is when I got issue 8 and then that's it. Since then we have had teased new issues, every year or so someone posts on social media "oh there's a new one coming". There is a lot of media like this but for me, this is the one I think of most often as "what-if". I think it is a genius idea that is especially relevant in today's world and I am sad that we may never get another one.

So I was expressing this to a boomer friend who casually responded with "why don't you just make your own"? LLMs allow us to generate lots of artwork like this now, you can maybe cobble together a story, wouldn't it be fun to make your own issue for your own personal consumption? To him, this is the perfect use of this technology, allowing a fan to make more of the thing they love. To me it feels like if someone said "if you are hungry, there's a puppy over there, just pop that bad boy in the oven for 2 hours and have yourself a snack".

To me that's not a tribute or the act of a fan. It's making a forgery. The thing has no value to me as a reader if I removed it from the entire creative context of its creation.

What value does a forgery have?

There's a great short essay by the philosopher Denis Dutton from 1979 called Artistic Crimes. It's about forgeries, not AI, but every paragraph rings like a bell for the argument I keep having. You can read the entire thing here (it's not that long I promise).

What I was struck by was how many parallels you can draw between the two issues. Like AI art, the existence and occasional acceptance as real artistic works of forgeries has a special power to discredit the art critic and historian community. The success of a forgery creates a belief that the entire criteria by which art is judged by is fundamentally flawed. If something is amazing before you knew it was a fraud and trash after you came to find out it was fake, then the assessment of art has no basis in aesthetic properties. It is pointed to as a demonstration that the only reason a "masterpiece" is valued over a random painting at the flea market is that a famous person made the first one. Effectively all art is bullshit and LLM slop/forgeries are equally valid because they invoke the same emotions in the audience before the deception is revealed.

Now everyone, or at least everyone whose opinion I value, can agree that knowing if a piece of art is real of fake matters in terms of both its monetary value and its value historically. Even the most staunch Boomer I know would admit that fake paintings shouldn't hang in museums next to real ones. Where the two groups diverge, in LLM art and forgeries, is whether the assessment of its aesthetic merits should not be impacted by its status as real or fake.

Part of what is impressive about the performance of making something is that you are celebrating how much a person or group of people succeeded. This is why we delight in stories like "Stardew Valley was made by one person in his house for years with no money hand-drawing each piece of art in the game, so intense was his love and commitment to the project". You don't have to know that to enjoy the game, but pretending it changes nothing about how you enjoy it once you do it is nonsense. Same with movies like "Good Will Hunting". Sure it's a good movie, but the story of two childhood friends making it for not a lot of money and one of them leaving Harvard to shoot it changes your connection to the material. It can't not. You love the thing, you learn more about the human story behind the thing, you love the thing even more.

Part of the magic of wandering through a museum is that I cannot do the things the artists here have done. I cannot make modern art, I don't have a vision like those people do. I don't have the strength of character and conviction to commit to something like they do. I frankly lack the fucking spine to stand there next to a creation of mine that a thousand morons a day will look at and say "I could do that". No you couldn't and that's part of why its impressive to look at.

Forgeries, like LLM art, are misrepresentations of achievement. You want the same credit and respect as someone who actually did a thing, but you haven't done the thing. Once that discovery is made or, frankly, once that is suspected, the relationship to the work has been permanently changed. I'm no longer viewing it from the perspective of the output of a human who I want to succeed but instead skeptically as someone attempting to steal credit for work they didn't do.

The question of efficiency

The other argument the boomers make is about bottlenecks. My favorite comic isn't finished because the author is busy and human. But we could feed the machine the premise, let it crank out pages, have him touch them up, and, bim bam boom issue nine, ten, eleven, a hundred.

Let me be clear that I'm not precious about everything in my life. I am not one of those people who thinks everything must be handmade by candlelight. I love a Dunkin Donuts coffee. I love the ridiculous bucket-sized cup, the sugar content that could kill a horse, the milk-adjacent liquid that has clearly never seen the inside of a cow and is shelf-stable until the heat death of the universe. I love standing in line behind two guys in Carhartts complaining about a foreman named Chris who is, and I quote, "a real fucking prick, man." I could eavesdrop in an IHOP for the rest of my natural life and die happy.

Sometimes I want the $22 burger. Sometimes I want the McDonald's cheeseburger, no pickles. Both are fine. Both have their place.

But here's the thing the boomers keep missing: I get to pick.

The choice is the thing. And the AI rollout, as currently constructed, is designed to remove the choice. Not offer a new option but replace the existing one. The boomer argument, stripped of its language about productivity and democratization, is really this: we're going to make the cheaper thing, we're going to stop labeling it, and you're going to keep paying the same price because what are you going to do, not consume media?

Because it is cheaper and faster to make doesn't make it something anybody wants. My favorite local bakery makes a great sourdough that I try to get whenever I can. If they sell out, I don't buy a loaf of Wonder Bread and pretend it's the same thing. I go home without bread. Sometimes the honest answer to scarcity is going without.

This is what boomers can't hear. They think we're arguing about the quality of the output, and so they keep showing us better and better outputs as if that will close the deal. But we're not arguing about the output. We're arguing about what it means to receive a made thing from another human being. Every comic in my pull list, every trashy Space Marine novel, every weird indie game are letters from strangers who cared enough to spend years of their lives making something for me to find. The value isn't in the artifact. The value is in the fact that somebody bothered.

A machine cannot bother. It can only produce.

I would rather have eight issues of The Black Monday Murders and the ache of never getting a ninth than a thousand AI-generated issues that continue the story forever. The ache is part of the art. The waiting is part of the art. The possibility that the thing I love might never be finished is part of what makes it real. Forgeries don't ache. They just fill space, like the paintings in a Best Western lobby, hung there because a wall needed something on it.

All hail God Mammon. He always did prefer the cheaper option.

Midnight Train to Stockholm

July 03, 2026 9 min read in Personal

I was recently summoned to a meeting in Stockholm, a city I had somehow managed to avoid despite living in Copenhagen for years. My Swedish experience, up to this point, consisted entirely of trips to Malmö — the closest Swedish city to Denmark and, more importantly, home to a Costco. As an American living abroad, I am duty-bound to report there every six months so the proper authorities know I'm still alive and to procure my ceremonial barrel of peanut butter pretzels. It's less a shopping trip than a consular check-in.

Stockholm, it turns out, is much further from Copenhagen than anyone lets on. My options were to fly or take the train. Flying is technically a one-hour affair, but to make a 9 AM meeting I'd have to wake up at 4 AM, shuffle through security in a fugue state, and land in Sweden looking like a hostage video. I wasn't sure when I'd nap. This seemed insane.

Then I had what I believed, at the time, to be a genius idea: the overnight train from Malmö to Stockholm. I'd sleep en route, wake refreshed, stride into my meeting like a man who understood something about life that others did not. Plus, I love trains. Sweden has a high-speed line that does the run in four hours, but the night train takes its time, which sounded charming. It was not charming.

The first surprise came at booking. There are three tiers of experience: a seat, a couchette (whatever the fuck that is), or a private sleeping compartment. Since I wasn't paying, I chose the private compartment. This would prove to be the single smartest decision of my adult life, possibly of anyone's adult life. If you are reading this from some point in the future and you are over the age of thirty, book the private compartment. I don't care what it costs. Sell a kidney. I'll explain.

Departure

My train left Malmö at 10:30 PM. The station there is depressing in a way that's hard to articulate in that nothing is obviously wrong. There's a grocery store. There's a convenience store. And yet everyone inside looks stranded, as though they've been waiting for something that isn't coming. Small children roam in feral packs. There is a pervasive sense that this is the last train out of somewhere very bad, and that whatever is chasing everyone is still, perhaps, on its way.

Amtrak in the US has this feeling, a vibe that you are running from the law. I remember when my family used to take the train from Ohio to New Jersey, waiting for it at a train station that was basically a concrete bunker in the middle of a corn field. The concrete box would either be freezing cold due to too much AC or boiling hot due to too much heat. I would stay up late on the train and watch as parents would abandon sleeping children to jump off at stops and catch a smoke. They all had a nervous desperation that these Swedish travelers shared. It's the kind of place where you wouldn't be surprised to see someone take a SIM card out of a phone and throw it in the trash.

I stopped at the grocery store and loaded my backpack — my only luggage — with provisions: two large water bottles, wet wipes, a change of clothes, a bag of nuts, and, as an emergency measure, two Red Bulls, in case sleep failed me and I had to power through the following day on hatred and that cursed, faintly urine-themed energy drink. Then I stopped by the men's room, which featured a decorative fish tank whose sole occupant had a full-frontal view of the urinals. If reincarnation is real, Henry Kissinger is in that tank.

European train station bathrooms are often weird, but this was up there. First there was no automated system to get in, it was just a guy with a credit card reader. Also they were piping in tropical sounds to the bathroom which I assume is to cover the unspeakable horrors happening in the stalls. I felt uncomfortable that I kept looking at the fish and found it to be always staring at me. The tank was in really good condition, with incredibly clean water. I couldn't help but think maybe its better for the fish to die faster than live their entire lives staring at an endless line of men peeing.

I found my train and boarded, and I knew immediately I was in trouble.

This was an old-school train — varnished wood, worn blue upholstery, late-70s energy throughout. The regular seats were hard, upright benches with little fold-down wooden tray tables. They did not recline. Not a little. Not at all. Which raises the question: why call it a night train? Night train implies, to me, that at some point during the night, someone might sleep. But the seats also had bright lights above them that never turned off, which transformed them from sleeper seats into something closer to interrogation chairs.

The couchettes turned out to be stacked bunks — men-only, women-only, or mixed and the passengers were packed together so tightly that the gender segregation began to make a grim, practical sense. I'm not squeamish around strangers, but we're talking well within reach-out-and-stroke-someone's-hair range. My private room looked roughly like a jail cell: a cot, a light that turned off, a door that locked. In other words, everything I have ever wanted from a hotel. That's not sarcasm, I'm easy to please.

Naturally, I was far too curious about the rest of the train to actually sleep, so I set out to wander.

Train to Hell

The three conductors on duty were all wearing bodycams strapped to their chests, which is always an encouraging sign in that it suggests both that they had been attacked and that they had, at some point, done some attacking of their own. They were also wearing shorts, which felt deeply wrong. There's something unsettling about a train conductor without pants. It's a formal job. You don't want your pilot in flip-flops and you don't want your conductor showing knee.

I don't know what it is about shorts on men specifically that come across as clownish, but there is a ranking of jobs where one shouldn't wear shorts all the way to one cannot wear shorts. Doctors, pilots, lawyers, accountants are all pants jobs. Train conductor felt like a no-brainer that it would be a pants job, also frankly I think they should also have to wear a cool hat and have a pocket watch. In the same way I would bristle at a judge sentencing me to death in a Hawaiian shirt, a train conductor in shorts checking my ticket feels wrong.

I made my way a few cars down to see how the general population was faring.

The door slid open and I was hit, physically, by the smell of cheap vodka. Before me stretched a sea of Swedes, each with the specific facial expression of a person who has just realized they have made a terrible mistake and cannot un-make it. Two people were openly weeping. Three others were borderline-homeless-looking punk kids dressed exactly the way punks dressed in 1994 — I don't know who is still manufacturing M65 field jackets and military jump boots, but they're clearly still moving units in southern Sweden.

One of the punks was vomiting into a plastic bag, seated next to a very sweet-looking, deeply concerned young woman who had presumably boarded this train with hopes and plans. She had the look of a woman who had her life together. The conductors arrived, spoke to him, were told to fuck off, and then quietly relocated the young woman to a new seat the way you'd move a house plant away from a leaking radiator. The punks then began joking loudly among themselves, two of them taking turns retching up what smelled like vodka cut with unleaded gasoline.

God help anyone trying to sleep back here. Between the puking, the reek, and the punks openly hitting on every woman within shouting distance, it was like being trapped on a Greyhound bus that had sworn a blood oath never to stop. I watched a man roughly my own age attempt to sleep by laying his face directly on the wooden tray table, earplugs jammed in, arms limp at his sides, in the international posture of I have given up.

I left when a boyfriend and girlfriend began fighting because she had proven surprisingly receptive to the advances of a punk kid whose body odor was strong enough to reach me three rows back. The boyfriend — sitting directly next to her — took issue with this, which seemed reasonable. I moved on to the meal car.

The meal car was the hangout, the refugee camp, the place where people who had discovered they couldn't sleep in the bunks came to sit and stare into the middle distance. Two young women were seated at a table nearby, one of them work-shopping, in English, why she deserved better than her current boyfriend in Malmö.

"I don't think I should settle for average."

Her friend was being supportive and kept trying to inject something about her own life, only to be steamrolled every time.

"Yeah, I know exact—"

"It's just, I work so hard at school and he doesn't."

"My last boyfr—"

"Maybe when we get to Stockholm we go buy some nice dresses and go dancing."

"That sounds fu—"

"Because I really do think I deserve to feel beautiful."

After a few rounds of this, I got bored. It was all early-twenties drama, and I don't say that with contempt because it's a phase we all pass through. In your twenties, you discuss your plans and feelings as though they matter, because to you, they do. In your late thirties, you come to understand that nobody actually cares if you live or die, and you learn, mercifully, to keep it all to yourself. It's one of the small gifts of aging, along with knowing how to fold a fitted sheet and no longer pretending to enjoy helping people move.

Friends in your 20s are your therapists, your closest confidants and your relationship counselors. In your 30s, you have an actual therapist and don't have to burden the people around you. At some point in everyone's life someone they love and respect will put up their hands and say "alright ENOUGH" and you'll realize how tiring you are. These women hadn't gotten there yet, but I did think the friend should start charging for this therapy session.

I retired to my cabin and fell asleep almost instantly. The rocking of the train, the smug satisfaction of not being propped upright in a wooden pew next to a vomiting stranger, and a modest dose of melatonin combined into something close to bliss.

Stockholm

We arrived on time. I grabbed my backpack and stepped out into Stockholm Central Station at 6 AM, which was nearly deserted. I found a coffee shop, ordered a coffee, and was charged an amount that made me wonder "should I open a coffee shop in Stockholm?". As he handed it over, the barista said, cheerfully, "It's my first day."

"You shouldn't tell people that," I replied.

I then felt terrible about it for approximately one hour. He, for his part, seemed entirely unfazed, or possibly hadn't heard me at all, which somehow made it worse. I carried the guilt with me through security, out onto the street, and into the cab I hailed for the twenty-minute ride to my meeting.

The meter began climbing almost immediately, and with a kind of enthusiasm I hadn't previously known meters possessed. He asked where I was from. I said the U.S. He said he loved Americans. He said Americans were the best people which, frankly, nobody says unless they're being tortured by us. He asked if this was my first time in Stockholm. He asked what I did for work. Every question was warm and generous and I understood, dimly, that I was being courted.

But it was only twenty minutes, how bad could it be? Eleven hundred kronor later, I stepped out onto the curb while he tried to press his business card into my hand so I could call him directly for future rides. Having just paid roughly $120 to travel the distance of a decent jog, I now understood his enthusiasm. He'd hit the jackpot, and the jackpot was me.

The meeting wrapped in a few hours. I took an Uber back for a quarter of the cab fare — a small, petty vindication I savored the entire ride — and spent the next four hours walking around Stockholm. It's genuinely lovely, and denser than I expected. Copenhagen feels like a city that was designed by someone who liked people; Stockholm feels like a city that was designed by someone who respected them but wasn't sure he wanted them over for dinner. More cars, less green, wider streets, harder edges. Every Swede I passed looked as though they were on their way to something slightly more important than what I was doing. It was beautiful in the way certain people are beautiful — the kind of beauty that doesn't especially need you to notice.

Conclusion

Should you take the overnight Swedish train? Honestly? Probably not this one, unless you are terrified of flying or being sober. The only tolerable option for an adult human is the private cabin, and at that price you can usually just fly. But it was, undeniably, an experience and one I would happily repeat, provided someone else were footing the bill.

TL;DR

Setup

Lesson 1: Don't handwrite the filters

Lesson 2: Sort keys matter sorta (dad joke, you are welcome)

What does ORDER BY mean in ClickHouse?

Why is timestamp in there twice?

Lesson 3: Projections are trickier than I thought

Lesson 4: Grafana ClickHouse Plugin has a lot of sharp edges

Conclusion

What are the high-level problems with windows management in 2026?

What have people already tried?

Graveyard of Attempted Solutions

What might a solution look like?

Reasons this design sucks

Fun experiment

How do you recharge your batteries in hell?

The Cursed Banner

GDPR Day

Why Did GDPR Spread so Quickly?

Japan and the EU

The US and Surveillance Capitalism

What does surveillance capitalism look like?

Double Movement

How does OpenTelemetry Work

Process of adding a new feature

Attempting to test it

Conventions about Semantics

Potential Solutions

The Smell of Ink and Nerds

What value does a forgery have?

The question of efficiency

Departure

Train to Hell

Stockholm

Conclusion