🎉 Introducing `refactorlah` CLI -- fully open source. Agents are surprisingly bad at renaming/moving files. They are doing `git mv`s, then dozens of `grep`s, then hand-editing imports/namespaces/references while burning tokens and wasting time on tiny refactors. The `refactorlah` CLI gives them one boring command instead: `refactorlah move A B` or in bulk `refactorlah move --use-list A,B tests/A,tests/B` Then file/s are moved, and namespaces, imports, references, etc. automatically updated. Ambiguous renames are skipped and reported, so the agent can follow up manually instead of guessing. Support for Go, Python, PHP, Symfony, Twig is landed. A PR for JavaScript, TypeScript, Vite etc. is open. I built and dogfooded `refactorlah` over the last few weeks. It works better for me than keeping an MCP server running in an IDE or making agents struggle through long tool-call chains for basic moves. Still in alpha! Experimental, built with my best buddy Codex, and still dogfooding. More eyes -> more dogfooding -> releasing early. What do you think?

On a related note, I am not surprised that European politicians hammer all their thoughts into American AI -- probably everything, including all the classified stuff. Especially Ursula.

The feeling when you gave your machine some care, and ended up with 600 GB more unused disk space. 🥳

I first thought I am missing something, but Codex -- on macOS -- for detached chats, really dumps a Codex directory into `~/Documents`, which is iCloud synched, and there is no way to change it. 🫠

Quite digging Tahoe. It has some glitches and weirdnesses, but overall I think I should have updated earlier.

"Given that people can write dangerous software on macOS and Windows, the U.S. Department of Commerce has ruled that, going forward, foreign nationals are no longer eligible to use products from these companies."

Finally updated to macOS Tahoe. First design impression us great.

Everything about Anthropic makes me feel uncomfortable. Absolute bitch move.

How did we go from the USA being the greatest country in the world to being a national security risk for every other country, and a supply chain risk for every business outside the US? They govern by moods; which makes them no different from any third world country with no legal certainty.

Okay, problem solved. 😂😂😂

Yann LeCun headquartering in Europe will pay off big time.

Called it! Just didn’t expect it to happen with an already released model. The rest of the world has to realise that the USA nowadays is the adversary, unfortunately. So, how can we stop training “their” models for them? And how can me make them stop steal our knowledge?

Every now and then I need to duplicate a private GitHub repository, with all its settings and stuff. And then I remember that it isn't possible for whatever reason.

Do you think Anthropic execs propose their ideas to Claude and when Claude says “You’re absolutely right!” they just run with it?

The ads I get on here are absurd. What they want me to do? Inform me about some fine Saudi energy regulations?

Building with AI is so easy! The possibilities are endless. We now can create anything we want. Have an idea? Build it. Easy. All we have to do is spending some time, and burning some tokens. Easy. Here is what I have built in the last 5 months: - CLI for AI assisted HTML/CSS to Tailwind migration - Native macOS Markdown editor - CLI for managing agent/skill/prompts/tools - Social media data thingy - Government data thingy - Education data thingy - Monitoring tool thingy - CLI to make refactoring for agents easier Thing is, I shipped (or am using) only two of them. Because AI is great at getting the first 70% done, but the last 30% -- to make something actually usable and worth to ship -- are the hardest. And they will cost a tonne of time to get right, because we have no idea what the code does. The promise of being able to do everything "easily" with AI is a false promise, that is sold to us as the new dream reality.

Ladies and gentlemen, after ~5 months of procrastination, today, I made the first commits to the project I actually have to work on.

Do the people who let agents fire tonnes of SSH commands to their servers, to change and debug stuff, know that their servers are insecure?

I swear, me going to bed at 3% weekly limit left, dreaming about what to prioritise or whether to upgrade, then waking up to this is... Timing could not be better, haha. 🫠

Opening a project I haven't worked on for a few months and tell the agent "I forgot what happened last. Bring me up and running." is kinda priceless.

👏 Subscriptions subsidise themselves. It is called subscription economics. 👏 API prices are not provider costs, and say nothing about margins. 👏 Subscription usage multiplied by API pricing is not a serious metric.