📺 Mike Judge’s Beavis and Butt-Head – nearly the opposite of The Wire, if you’re of a certain age the new episodes of Beavis and Butthead continues to entertain in a very specific way. If you liked it as a teenager, you might continue to enjoy it today.

📺 The Wire – several episode in, I think this time it might stick. We are watching this show. 🤣 It’s slower-paced than crime dramas of the Breaking Bad era. But, the characters are deep and there is quite a bit more metaphor than a show might have today. So far, worth it.

For many models, particularly the largest ones, max effort is likely to try to solve an entirely different problem than what is asked of it. Ask it to make the button a little more blue and fix a layout problem, but the model goes off and reinvents hyphenation from scratch.

It’s like the phenomenon where a table of engineers is given the bill at a restaurant. Instead of splitting the bill by the number of people at the table, every engineer has to look at the bill and try to invent a way to split it up fairly and correctly. In the process, democracy is reinvented from first principles, thus wasting everyone’s time and energy while not even solving the right problem.

I had been using Sonnet for a while because Opus and Fable were a bit too eggheaded for me. But, Opus at medium effort turns out to be pretty good these days.

📚 The Power Broker – the short chapter about LaGuardia (temporarily, I presume) stymying a Moses power grab that follows the very long chapter about Moses treating his brother and sister awfully? That’s a great shot/chaser.

I’m into The Good Sentences now. Send me your favorites!

Madrid has nice things

10 observations, nice things, and little ideas that made Madrid pretty pleasant for a week-long trip last spring:

  1. You can’t make a city like Madrid with token spend. You probably can’t even make Madrid by gathering numerous friends in an underdeveloped grass plain and deciding you’re going to make a great city. 🤷🏻‍♂️
  2. Extremely related to the previous: Madrid has many nice things because it’s a lot older than most American cities. It necessarily evolved and was built at human scale. It was probably a major pain to get a (pretty good, IMO) subway underneath it. But, given the fixed locations of palaces, royal gardens, etc. you get a pretty pleasant place.
  3. Money invested in a city doesn’t automatically improve it. And it can’t buy you happiness. But it can avoid many kinds of misery! (Possibly only if you invest capital in making things nice, and invest socially to avoid misery? I know, it’s going to be difficult to get the private enterprise types to wrap their heads around that one.)
  4. It is quite charming to have parks, gardens, etc. spread throughout the city. Pocket parks and recreational parks are delightful, but something more akin to English or royal “gardens” is even better. Robert Moses should be jealous on this point. (He probably was not.)
  5. Good public transit makes every part of a city that it reaches 20% better. One system of buses, subways/light rail, and streetcars/trolleys is a delight. Obviously, the places that public transit reaches have to be walkable. Lucky for Madrid …
  6. Many things in Madrid are good simply by having evolved at that human scale. It’s challenging to spend or regulate into that. For better or worse, most American cities have car-scale deeply embedded in their structure. (🤔What can we do moving forward?)
  7. Counterpoint: the pall of smoking is prevalent the moment you can see the sky in Madrid. It is the one jarring thing we constantly noticed and thought we have it way better back home.
  8. Granted, Madrid seemed lovely at the center of its energy (and geography). The Disneyland of Madrid, per se. Like useable (not even good?) public transit, I’m not sure what percentage of Americans would ever actually experience this sort of thing. Even on vacation, seems like a rarity.
  9. Related, Madrid has a plethora of museums, history, and cultural touch-points. As cities go, you can only luck into that and then try to preserve it. A city unfurled across history is a rare delight.
  10. Gardens, useable benches, people watching. We could have these things in US cities if we wanted. And, were willing to address all the second-order forces that prevent us from having them now.

In summary:

  1. Madrid can have nice things because it’s old
  2. Except the smoking
  3. Walking amongst history, museums, culture: no notes

Therefore, enjoy Madrid like I did:

  • Go to museums, enjoy the history, see old things
  • Walk and take public transit whenever feasible
  • Drink Spanish vermouth, sleep in

📚 The Pine Barrens by John McPhee – I would not have thought I could enjoy a series on the ecology and peoples of an obscure yet famous, out of the way yet vast, part of New Jersey. Or such a place would be the setting for idiomatic language, proposals for supersonic jet ports, intense but rejuvenating forest fires, requisite tall tales, and remarkably clean water. I guess that’s a point in McPhee’s favor as a writer!

As goes writing, so goes coding

Writing and structuring can be highly pleasant tasks. But switching from writing to structuring and formatting is like jumping between your favorite song and your favorite movie every few seconds. You ruin both. The more we enjoy doing one thing, the harder it is to abruptly shift to something else, even if we normally enjoy it, too.

— Separate Writing and Formatting

Separate the tactical stuff like new functionality, tests, design, refactors, etc. from the strategic stuff like design, research, brainstorming. Decide if you should remove some of that code (you just paid a few bucks to generate) in the first place.

I keep hearing writers going on about enjoying sentences as standalone objects.

A well-known writer got collared by a university student who asked, “Do you think I could be a writer?”

“Well,” the writer said, “I don’t know. . . . Do you like sentences?”

— Walking with Annie

We are largely past this for coding, for better or worse. But, there is still the holistic appreciation of how a program solves its problem space. Whether a human typed it or not, there’s something to a program that concisely composes levers and concepts out of language primitives and then uses them to solve many problems or one problem in-depth.

Fall is setting into the Pacific Northwest, and I’m (mostly) fine with that. Grey, rain, not particularly warm. “Stay-inside” weather. I prefer summer, but the changing seasons are nice. This, and adjacency to multiple biomes, is why we moved to Oregon!

In your face, Texas

On our last visit to California (outside the Disneyland bubble), I noticed more of the in-between places. Oregon’s in-betweens, at least along the western part of the state, are way more interesting than California’s between San Jose and Monterey. Grasslands remain not-my-thing.

🕹️ I’m trying to play fewer open-ended video games. I should select games that are designed for some idea of finishing and reaching a satisfying ending. But, Marathon and its (open-ended) skill-climb is often compelling. Possibly, I will never not play an activer Bungie game. Which works out, since it seems like their future is on the fence.

I need more close-ended, finish it, and you’re done games. Or, to replace gaming with something more fruitful and becoming of a grown man. I could be pretty good at a lot of things if I’d spend the dozens and hundreds of gaming hours at other hobbies. 🤷‍♂️

Things one might think about coding agents

Things were widely discussed, ideas were shared, I was convinced. Later, I read more opinions and experience reports, and I was convinced otherwise!

The discourse is sometimes wrong, no matter how intuitively right it seems. Superficial isn’t super-right.

Anthropic, OpenAI, etc. intentionally weaken their product

If you find yourself thinking, “are the models worse today?”, it’s possibly time to step back and look at how you’re using them. Which is not to say it’s a “skill issue”. More likely, your setup has drifted away from how the models are trained to work within their harnesses.

The models (probably) don’t vary much day to day, inference and harness bugs aside. You likely want to reduce the size of your prompts, skills, and CLAUDE/AGENTS.md. Let the agent do its thing without oversteering and verbal micromanagement.

That said, the frontier labs have been known to vibe code rigorously evaluate and release code changes which result in elaborate bugs that have generated worse results, poor reliability, increased costs, and/or all of the above. Sometimes, you gotta look at the status page and see if the carnage is on their end or yours. I do not like this subtle gaslighting-adjacent quality of coding agents, but seems like we’re stuck with it until the labs compete on robustness.

The big labs are diluting their models for operational/performance reasons

This is a restatement of the previous one in the context of the next one.

There are a bunch of compelling stories that might make you think the labs are intentionally diluting their models. Because they’re afraid of Mythos’ capabilities or OpenAI’s models escaping containment. For operational expense reasons like reducing cost of inference. Because they’re just dishonest, pathological liars.

It is again, unfortunately, more likely that human error, and human error magnified by coding agent support, are interacting in ways that make it seem like the models are consistently worse than they were last week or last month. But really, it’s good old-fashioned bugs compounding to the extent they plausibly resemble malice.

AI is toast once the subsidies (VC money) runs out

We should enjoy the discounted token party while it lasts, right? Hence, all the double-usage weekends and “xyz model is available on all plans for the next several days” promotions.

I’ve come around to the mental model that reduced token cost is merely one lever in a sophisticated pricing model and subscriber retention strategy. I’m of the mind that the token party will probably continue, if you don’t mind hitting a usage limit every once in a while.

That said, the labs increasingly charge per-token consumption pricing of more customers and use cases. If token subsidies are important to you, get good at using first-party CLIs like Claude Code. This is currently true for Anthropic and their incremental but obvious restriction of how subscribers can use their tokens. Seems inevitable that other agent vendors (OpenAI, Cursor) might go this way, despite the good will they’re accumulating right now in not doing so.

I’ll just tell the coding agent to never, ever do the Extremely Bad Thing

Determinism is (still) a thing you can only get by writing helper scripts in a deterministic language. Or, by trusting that the little box you’ve put your agent into really impermeable. (Which it seems isn’t even true for the most clever of laboratories.)

Apparently, there’s a magic word, “gates”, that matches the domain language of machine learning to invoke guardrails. I’m not sure how effective it is. Eventually, even text like “you must never get the gremlins wet” is just another bag of floating-point numbers to the model. They’ll eventually get the math wrong.

I will never learn a new thing again

Sounds like a personal choice to me.

Granted, this is a bit of a straw-man argument. The stronger form of this argument gets back to the essence of what software development, creative work, and many kinds of knowledge work really are: problem-solving searching for an outcome that your team has decided upon.

The puzzles solved along the way are often means and not the end. For a lot of us, there’s pride in knowing the solution to the old puzzles and bringing those ideas to bear against new puzzles. Working with AI bypasses the rote application, and that will always jostle some feathers.

That doesn’t mean an agent can’t help one learn to solve new puzzles or fit that knowledge into the gaps in your experience. Ask the agent to help you learn how to solve/write the thing. Instead of just writing the thing for you and worrying about skill atrophy. Or, ask it to help you get started and remind you about the syntax or functions to use.

The little gizmos are pretty good at putting together explanations. At checking them for hallucinations against what a database query, or linter, or a web search, actually say about an assertion. At putting that all together as a haiku, sonnet, or graphic novel.

Now is the time to either celebrate you never really figured out the syntax for shell scripting, with all its fiddly sigils, or ask an agent to make a bespoke tutorial for you to finally learn shell scripting and all its spooky action at a distance.

Will I solve the same kind of puzzles I did before? Possibly not. Will I solve problems as a craftsperson of expressing yourselves to computers and people? Probably not.

Am I going to learn things? Only as much as I want to. (Which is quite a bit.)

Watched: Slow Horses Season 6 📺 The show is as perfect as ever, in its own way. It’s the best show on Apple TV that you’re probably not watching. Horses has everything: flawed people, workplace comedy, spy thriller stuff. And it asks the questions not typical in the spy genre: who are the people behind the extremely well-coifed spies, how do they keep getting away with all this, what if they all became a kind of begrudging workplace family in the end?

Things I learned in autocross this year

Not my car or numbers, but I appreciate the feeling.

Use more gas and more brakes. Actually, just tap the brakes. It’s not a track, the speeds aren’t that high, the braking zones aren’t big.

Oregon Raceway Park is a fantastic track. It’s in Grass Valley, OR which is perfectly named. It’s a lot of grass, in a valley.

Talking to strangers about cars is half the point. The other half is mostly about solving a new course in a limited time and somewhat about picking up cones other people knock over.

Sometimes, strangers and friends alike will feed you before/after you drive, it’s hard to beat that. Don’t drink fizzy beverages immediately after standing in the sun without anything to drink.

Weather patterns in legit, ski-able mountains make no sense if you’ve spent most of your life in the plains of Texas. Whatever kind of variance in temperature and conditions you could imagine in a day, double it.

Autocross is a tires game. Better grip makes for greater confidence makes for better times. But, when they say the best upgrade for your car is a better driver, they’re absolutely right.

Finished: Plato’s The Republic by Plato 📚. Plato sure doesn’t like creative-types! Talk about zero self-awareness. Though, I suppose “philosopher” wasn’t considered the same school as sculptors and poets back in the day.

I had been feeling a bit two-dimensional. Coding agents, too many attention-hungry algorithms making my feeds a bit predictable. Maybe (finally) fallout out from video games. Distractions (apps and social media) too close at hand. Indicators of doom in our daily news.

I wanted to find more shape, variety, and liveliness in my life. Keep the coding, technology, writing, and reading. Add more of the city and novelty and exploring. Bring depth and pacing back to my life.

Spoiler alert, I have found some of it. And, the journey continues.

(I’m trying to unfurl an idea across blog posts so that it will all come together nicely for the next newsletter I write. I’m trying not to make a big deal of it and just get on with it. But of course, the idea isn’t emerging as a series of words in the punchy, crispy bacon way I’d like. So, more of an oatmeal of words, the sort that slightly burns the mouth, will have to do.)

Social media is where you’ll find the most sublimely interesting and entertaining people in the world. Right next to them are people with so many bad takes and shallow enthusiasms. Of course, all the clankers, some masquerading as human and some transparent in their toaster-ness.

You may think it’s where your people are, but the whole point is to mix your people with the Havers of Takes and the Obviously Artificial. And the advertisements and Calls To Action. You’ll come across your people every once in a while, in between frequent run-ins with people who can’t possibly think that (right?) and obvious ads claiming you’re missing this one thing in your life (click here to subscribe).

Of course it’s The Bad Place!

Some days, you write for yourself. Other days, you write for everyone. Most days, the world writes for you. 🤓

A weekend in Carmel, CA

My only goal was to spend the weekend by the pool. I did, and it was everything I’d dreamed it could be.

The big blur – Brian Bailey nailed it. Wish I’d thought of this framing.

We’re living through the big blur. Product, design, engineering, mobile, web, QA. Roles are fluid. Previously clear lines now fade at the edges like lines in an Agnes Martin painting.

I’ll yes-and this: not only are roles and activities blurring, but time is as well. A day spent discussing code in reviews, grilling a feature with an agent, steering another agent to focus on what’s important, keeping projects on track, providing feedback, and monitoring Slack channels leaves the brain feeling like a page full of great ideas all smudged out by the wet hands of an automaton toddler.

Once more, with feeling – another season, another newsletter. Trying to get a conceptual hold on how to feel human in this continually liminal time in software and technology. Current hypothesis: maybe thinking slowly and holistically, whilst acting quickly, is worth exploring.

We continue to feel all sorts of ways about machines generating code, writing on the web, composing images, deciding things that humans used to decide. Robots continue to feel nothing. I bet you can see the problem here.