something that keeps bothering me is how some people see slop as something that is happening to them, like they're some kind of helpless bystander powerless to do anything about it
excuse me but you are a software engineer, you literally learned how to make sand think and make computers do your bidding?! and now you can't do anything about it? please
the 2 things you should know to cut through all the noise:
1. you need to give your agent a way to verify its own work (i just posted about this so won't rehash)
2. your codebase is memory
the 2nd is not new either, we have always had lint rules, compilers with good diagnostics and strong type systems, LSPs, etc to help us write better code. there is no difference when agents write the code instead, you just need more of them, and stronger ones.
forget rules and skills. you can add those later.
yes, your codebase is a form of memory to agents! agents love extending what already exists. so every nit and every annoyance you've ever seen, every mistake and bad pattern you observe from agents, needs to turn into a constraint, or the agents will just keep repeating those same mistakes like a virus in your code. then you keep observing and repeating this process until your codebase is locked down enough that even agents with little context can one shot a high quality PR.
an even better direction is to refactor/rewrite your code such that agent anti-patterns become categorically impossible. use type systems to your advantage, use libraries/frameworks that reduce the architectural search space, or heck, create your own framework like we did for grok bot and design high quality primitives and well thought out APIs
I asked Codex (CLI) to figure out what was using up all my memory and it started the Activity Monitor UI and started filtering and killing processes from there.
Yeah Sol we know you're good at computer use, but you still know how to use the shell right...
It used to be very important to prevent main from going red to prevent a non-deployable main or context switching to quickly fix it. So you ran all tests on each PR and used a merge queue.
Now if our main branch goes red, an agent get's notified and fixes it. No merge queue.
Jacob Rask retweeted
recommended reading, though i still disagree, as explained on @syntaxfm early this year.
for code, both memory and docs are not efficient. your codebase is all you need. make it modular, make modules small, at most, have a little maintained map of where what is.
liao.gg/blog/agents-dont-nee…
Spawning out some tasks in Pi to Qwen, it’s plenty fast on my MacBook Pro with 48gb ram.
Figure I might as well use it for resolving merge conflicts and creating PR descriptions, saving on subscription tokens.
Jacob Rask retweeted
If you want to play around with an early thing: you can now give us money and we give you tokens in return. radius.earendil.com/
A nice way to make a large codebase stricter over time: add a lint:strict command, use AGENTS.md to get agents to run it on new or significantly changed code.
This used to require either changed-file linting and time-consuming cleanups, or massive changes across the board.
You give your agent n lines of instructions or skills, but you also give it thousands+ of lines of code as implicit conventions. You might not read the code, but your agent sure does!
If your agent misbehaves, maybe you don’t need a new skill, just a proper refactoring session.
Tried Opus for 5 minutes, immediate Claudish
"That's a better argument than the one I was making, and it flips the recommendation. Prose is a suggestion; a type error is a wall the agent has to walk into."
I’m leaving Volvo Cars after 7 years! I had a great time.
The monorepo I formed now hosts all of volvocars.com (Next.js + microfrontends), with 400 committers over the years
The design system I created now has massive adoption all the way from .com to factory floors
designsystem.volvocars.com/w… really happy about the new Volvo Cars Design System docs!
The “light dom” (CSS) slots APIs work really well designsystem.volvocars.com/w…
90% of my Claude usage is now to get `claude -p` reviews from Opus from within Pi (with GPT 5.5). Guess I'll cancel my plan.
Using Pi really changes the way you think about software. Anything you want it to be able to do, you just prompt it to change itself.
Some people are attributing this 8-components-per-line stuff to React but I think it’s a fundamental problem with UI component maximalism in general
Sometimes a component is not the right abstraction
Event delegation and CSS cascade are tools you can use *with* React
Replying to @github
After overhauling the React architecture of GitHub's "Files changed" tab, we dropped Interaction to Next Paint (INP) from ~450ms to ~100ms.
📉 React components cut from 8 to 2 per line
🔑 O(1) data access with JS Maps
🪟 Window virtualization for giant PRs
I toggled a feature flag today and 30+ web apps across 50 countries got a new theme
volvocars.com









