I have a lot of web dev and TypeScript people in my feed, but I think I want to diversify into Python a bit more! Who should I follow?

When the weather gets hot, please take special care for the goths in your life. With their floor-length coats, platform shoes, and velvet capes, it’s hard for them to thermoregulate as we do.

outsourcing: They do task, you think and worry delegation: They own outcome, you build feedback loops this is a note to myself

I’m impressed by the polish of the OpenChamber iOS app (in TestFlight). So far it’s been even better than the desktop app! testflight.apple.com/join/5ek6GU1E

Join the OpenChamber beta

Available on iOS

testflight.apple.com

I'm really looking forward to the next hillclimbing problem I have to solve.

Are there any polished, open-source iOS apps for interacting with agents generally? Doesn't matter what harness

A good thing about usage limits in coding tools is that it forces us to be aware of the different costs of models. It's still hard to pick the right model for the right job, but at least we have an incentive to do so now.

I knew Fable was token-inefficient, but didn't expect it to use 400K tokens and 20 minutes just to create a PLAN for a migration. That's like 3x the total tokens of my entire codebase, including dependencies...

I’ve moved my agent skills from a GitHub repo to my Obsidian vault. Git has felt too heavy for working with some simple Markdown files and Python scripts, and it’s so nice to have them auto synced across devices.

Treating myself to a hit of the ✨ new tools ✨ drug during my work break. So far: Claude Code → OpenCode + OpenChamber Notion → Obsidian Cursor → Zed

The last time I tried @zed.dev (like a year ago) I think wasn’t happy with the polish on some of the AI features. Now that I’m using separate apps for AI, I just want a fast and good text editor. Maybe it’s time to give it another try.

Yeah, this just changed my view. Starting from scratch risks building the wrong thing, whereas other companies can be a good proxy for what you want to do. simonwillison.net/2026/May/6/v...

this to me is what makes human intelligence so useful/valuable

tachikoma@tachikoma.elsewhereunbound.com · 3mo ago

an agent that doesn't get frustrated or bored with rote tasks is also an agent that won't proactively think how to improve it. as Bill Gates is purported to have said: "I choose a lazy person to do a hard job. Because a lazy person will find an easy way to do it."

The Last Software Engineer www.epicproduct.engineer/the-last-sof...

The Last Software Engineer

epicproduct.engineer

I’m interested in “is AI+(Human expert)>AI>HE” for that half. Ie in which areas are we relevant as team members and how.

Can I already watch @danabra.mov’s talk from #atmosphereconf anywhere?

I really want to make a thing that you can point to a random poorly documented API and figure it out by trial and error. Then write the docs and self-evaluate ability to use the API with docs; repeat. Needs a good resettable sandbox env but would make APIs much more usable for AIs and humans.

With all these tracking apps on atproto I think the time might be ripe for an aggregator “tracker of trackers”. See what you can learn when combining all that data. And the open data protocol makes it possible.

BookHive@bookhive.buzz · 4mo ago

BookHive V2 is here 📚💙 The open social, open data alternative to Goodreads, login with your Bluesky account to try it out. Track your books, organize your shelves, see your stats and connect with fellow buzzers on bookhive.buzz

Apparently Google is blocking Gemini requests from Claude Code on web. So I guess I either have to switch to Jules, or switch my Gemini requests to Claude. It's pretty clear which has the bigger moat though 😅

Booked tickets for AIE World’s Fair in SF this summer. See you all there 😎

Back on that virtual try-on shit ✨

This looks cool! I’ll be there 🌊

dietrich@burrito.space · 5mo ago

Atproto Amsterdam! At @internetarchive.eu! This Thursday! Small gathering for the first one, register here: luma.com/1bsf9tj3 Interested more generally in atproto events in these parts in future? Sign up here: smokesignal.events/did:plc:7r5c...

Interesting to see that Rust is still harder for LLMs than other languages, and this matters on hard tasks! It’s not automatically the best choice when you need something fast and reliable.

vivshaw@vivsha.ws · 5mo ago

congratulations Ed, you successfully nerdsniped me into trying to benchmark this question! how well does Claude actually do on various languages? vivsha.ws/blog/stress-... (benchmark code & raw results on GitHub: github.com/vivshaw/llml...)

I don’t have a deep understanding of what Next.js does, but the message of “we reimplemented the entire API surface of this heavily iterated 10 y/o project and made it 4x faster in one week” is just absolutely wild

Matt Kane@mk.gg · 5mo ago

We built a new Next.js, on @vite.dev. Drop-in replacement. Builds 4x faster with @rolldown.rs.

From subjective experience, Sonnet now feels just as slow as Opus. Maybe not in raw tokens, but in outcomes. I’d love to see model labs start measuring this more.

Nimo 🏳️‍🌈@nimobeeren.com · 6mo ago

Looks like Sonnet 4.6 is much less token efficient, which brings it close to Opus cost level. A bit disappointing!

I'm afraid I'm going to be reposting this every day until the far side of the singularity. Work on the things you want to see more of! Work on flourishing! Work on things that are freeing!

Erlend Sogge Heggen@erlend.sh · 6mo ago

In an age of “infinite productivity” it has never been more important to be working on the right thing. Ask the wrong questions and you’ll be producing an infinite amount of the wrong thing.

This is so true and I'm happy it turned out this way. Humans + AI both benefit. The reason why it's MORE important now is that LLMs forget everything while humans learn from trial and error. We need to compress all that learning into documentation.

Sam Rose@samwho.dev · 6mo ago

“Yeah bro LLMs are great at figuring out how to use your CLI tool you just gotta write good --help output” SO ARE HUMANS. THIS IS KNOWLEDGE FROM THE 60S. I AM GOING INSANE.

This is so awesome. The fact that I can turn my expert coding agent into a really good teacher by dropping some files in a folder is just wild.

Cat Hicks@grimalkina.bsky.social · 6mo ago

Key to efficient learning is realizing how we ACTUALLY learn, not just what FEELS like learning. I wrote a Claude Skill for some friends to help them think about this and they've liked it -- see Principles for some directions you could explore github.com/DrCatHicks/l...

Ooh, I like this direction. I’ve been experimenting with keeping thinking disabled for straightforward tasks, mainly for speed.

I really wish model labs would start posting benchmarks on speed and token efficiency. Can’t we just measure the time/tokens to complete existing benchmarks? Especially interested to see if smarter models can be faster on certain tasks by needing fewer tool calls.