Logs
This was cross posted to X, see the thread.
I built a coding agent back in 2022, 2 months before ChatGPT launched:
This was cross posted to X, see the thread.
People like the farmer analogy for AI.
This was cross posted to X, see the thread.
It is clear at this point is that GitHub’s trust and data models will have to change fundamentally to accommodate agentic workflows, or risk being replaced by other SCM
Anthropic earlier last year announced this pricing scheme
This was cross posted to X, see the thread.
I believe a “Christmas of Agents” (+ New Year of Agents) is superior to “Advent of Code”.
I finally brought myself to develop certain features for this blog which I wanted to do for some time, having a button to toggle light/dark mode, being able to permalink page sections, having a button to copy page content, etc.
Update (2026-09-18): Anthropic has finally yielded, 375 days after I posted this. Claude Code 2.1.277 now falls back to AGENTS.md when there is no CLAUDE.md in a folder:
We’re adding support for AGENTS.md to Claude Code.
Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md.
You can toggle this behavior in /config.
AGENTS.md support is built off of Claude Code mods, our upcoming way to customize the Claude Code harness.
This is a built-in mod, but you’ll be able to build custom versions of project instructions yourself as you’d like too.
You can see the source for the mod here!
— Thariq Shihipar, 18 September 2026
We do not need this workaround anymore. Keep the
AGENTS.mdfiles and delete theCLAUDE.mdsymlinks. Theclaude-md-symlinkerbelow was only ever a patch for a gap that Anthropic has now closed.
uv is now the de facto default Python package manager. I have already deleted all pythons from my system except for the one that has to be installed for other packages in brew.
I started using Claude Code on May 18th, 2025. I had previously given it a chance back in February, but I had immediately WTF’d after a simple task cost 5 USD back then. When Anthropic announced their 100 USD flat plan in May, I jumped ship as soon as I could.1
-
I previously had the insight that Claude Code would perform better than Cursor, because the model providers have control over what tool data to include in the dataset, whereas Cursor is approaching the model as an outsider and trying to do trial and error on what kind of interfaces the model would be good at. ↩
Dwarkesh Patel has recently interviewed Sholto Douglas and Trenton Bricken for a second time, and the podcast is very enlightening in terms of how the big AI labs think in terms of their economic strategy:
Item #: SCP-3434
Anthropic has just released a GitHub Action for integrating Claude Code into your GitHub repo. This lets you do very cool things, like automatically generating documentation for your pull requests after you merge them. Skip to the next section to learn how to install it in your repo.
Certain types of work are best done in one go, instead of being split into separate sessions. These are the types of work where it is more or less clear what needs to be done, and the only thing left is execution. In such cases, the only option is sometimes to work over the weekend (or lock yourself in a room without communication), in order not to be interrupted by people.
If you are a developer, you are annoyed by this. If you are a user, you were most likely guilty of this. I am talking reporting that something is broken, AND deleting it.
One common thing about sports noobs1 is that they don’t warm up before and cool down after an exercise. They might be convinced that it is not necessary, and they also don’t know how to do it properly. They might complain from prolonged injuries, like joint pain.
-
Including me before I started to receive proper training. ↩
This post had some discussions on Hacker News, see the thread.
Satya Nadella, shares his thinking on the future of knowledge work (link to YouTube for those who don’t want to read) on Dwarkesh Patel Podcast. He thinks that white collar work will become more like factory work, with AI agents used for end-to-end optimization.