Notes

I’m noticing an interesting branding trend among software companies: a shift towards something more nostalgic and warm. It started with serif fonts and warm colors but AI-generated images is making it a bit broader.

Amp Code is going for a sort of neo-romantic mythical feel. Retool is inspired by 90s anime backgrounds. Radiant has retro computers in cozy offices with nature views.

All of them signal something human, calm, contemplative — seemingly trying to stand out from the mass-produced / boring / sloppy feeling of AI-generated average-looking websites. Reminds me of Beyond The Black Rainbow, Project Cybersyn, and Soviet-era space posters.

Anti-patterns: things to avoid

If you open a PR with hundreds (or thousands) of lines of code that an agent produced for you, and you haven’t done the work to ensure that code is functional yourself, you are delegating the actual work to other people.

I agree – it’s rude to hand off AI-generated code (or writing, or business proposals, etc.) to your coworkers without doing the real, difficult work of verifying and understanding it first. I trust my coworkers a lot more than I trust an LLM.

I can imagine a future where this is no longer the norm, if there’s a collective agreement on the team that AI-generated code is consistently good enough, a Level 4 world where we are all PMs. Of course, there might be unintended consequences to outsourcing your entire team’s understanding of your codebase, who could say?

Which web frameworks are most token-efficient for AI agents?

Minimal API web frameworks are far quicker and more cost effective for agents to work with.

[…]

In terms of more fully featured frameworks SvelteKit and Django really shine - this doesn’t surprise me as they’re both extremely well thought through web frameworks.

A 2.9x token gap doesn’t matter much on a single task. It matters a lot when agents are building and modifying code hundreds of times a day.

Django, Express.js, etc. do pretty well but do not spark joy for me. My beloved Phoenix comes in last (💔). SvelteKit hits the sweet spot in terms of token efficiency and fun-to-use.

I wouldn’t pick a framework based on this metric, but this may become increasingly important over time!

Technically Your Name Is On It

Here’s a new project I built. Log in with GitHub (read-only) and it’ll tell you what percentage of your commits were co-authored with AI.

I’m at 32% over the last year (though this is slightly undercounting, since Codex doesn’t add the Co-Authored-By trailer to commit messages). Try it and let me know what you’re at!

Current

Current has no unread count. Not because I forgot to add one, or because I thought it would look cleaner without it. There is no count because counting was the problem.

The main screen is a river. Not a river that moves on its own. You’re not watching content drift past like a screensaver. It’s a river in the sense that matters: content arrives, lingers for a time, and then fades away.

New RSS reader from Terry Godier. This was an instant purchase for me. I’d been considering prompting my way to an RSS reader that worked just the way I want, but this is far more intentional than anything I could’ve come up with.

Burnout is Breaking a sacred pact

This post uses Jonathan Haidt’s elephant/rider framework: the elephant is the hedonistic, reward-seeking id, and the rider is the rational ego/superego that guides it.

It’s easy to extend this framework to explain burnout. You can think of the rider and the elephant as having agreed to a sacred pact: In exchange for doing what the rider asks, the elephant is promised certain rewards. When things are going well, the needs of both rider and elephant are satisfied, even if the balance isn’t exactly even day-to-day.

Burnout results when the rider asks the elephant, over and over again, to commit a tremendous amount of energy to a task, but then fails to provide the reward the elephant is expecting. As a result, the link between effort and reward breaks for the elephant, with catastrophic consequences for the rider.

It’s a very good post and well worth reading. I subscribed to the author’s RSS feed!

Agent Psychosis: Are We Going Insane?

There appears to be some competition in place to run as many of these agents in parallel with almost no quality control in some circles. And to then use agents to try to create documentation artifacts to regain some confidence of what is actually going on. Except those documents themselves read like slop.

[…]

When I watch someone at 3am, running their tenth parallel agent session, telling me they’ve never been more productive — in that moment I don’t see productivity. I see someone who might need to step away from the machine for a bit. And I wonder how often that someone is me.

This mirrors my impression after learning more about the Ralph loop that everyone’s talking about.

I recommend the normie loop: tell Claude/Codex/whatever what to do, watch what it does and steer as needed, repeat.

David Foster Wallace on Entertainment

A model of life in which I have a right to be entertained all the time seems, to me, not to be a promising one.

Crypto grifters are recruiting open-source AI developers

This system relies on your celebrity target being dazzled by receiving a large sum of free money. If you came to them before the money was there, they might ask questions like “why wouldn’t people just directly donate to me?”, or “are these people who think they’re supporting me going to lose all their money?”. But in the warm glow of a few hundred thousand dollars, it’s easy to think that it’s all working out excellently.

The fact that this seemingly revolves around X accounts should be raising alarm bells right off the bat.