Advertisement · 728 × 90

Posts by nikomatsakis

Preview
AGENTS.md AGENTS.md is a simple, open format for guiding coding agents. Think of it as a README for agents.

I got some questions about what people choose to check into git when doing agentic development. Do you check-in AGENTS.md? .claude directories or files? Hooks? I also asked in rustaceans-ai.zulipchat.com -- if you do Rust coding with agents, you should join that.

3 days ago 8 2 3 0

Shocking.

4 days ago 1 0 0 0

The docs are good! Enjoying reading them.

1 week ago 4 0 0 0
Preview
GitHub - tokio-rs/toasty: An async ORM for Rust (incubating) An async ORM for Rust (incubating). Contribute to tokio-rs/toasty development by creating an account on GitHub.

I finally published Toasty (async Rust ORM) crates! I’ll be writing about it soon (and why it took so long). github.com/tokio-rs/toa...

1 week ago 59 8 2 1

Read it. Agree it is insightful. Certainly the UX of chat makes it seem like more than it is. And certainly what it is is quite astonishing and powerful. I feel like this is one of the gaps: we are still in the hobby computing days of AI, but people are trying to sell it like it’s a polished product

1 week ago 2 0 0 0

I’ll take a look. I’m inclined to agree with the premise as you summarized it, just not sure what the better alternatives are.

1 week ago 1 0 1 0

Part of what sparked this was that I've noticed that reading Claude's thought processes ("I should try this, no wait, that won't work, how about that, that doesn't work either, let me back up, ...") can make me anxious. It is rather familiar from when I'm not operating from self, in IFS terms.

1 week ago 2 0 0 0

Anyway, mostly college-dorm-level musings here. I guess I'm thinking, I feel like the "talk to an agent" form factor is not obviously "right" to me. Of course we should go for more interesting ways to talk, like PR reviews, but I also wonder if there's a different general approach.

1 week ago 2 0 2 0
Advertisement

My cousin also told me once about an AI ethicist's theory that empathy is a scare human resource, and how it's unethical to devise technologies that "consume it". I'm not sure I buy that premise, I feel like the more you practice Empathy, the more you have, but I'm not sure how agents fit in.

1 week ago 1 0 1 0
Kate Darling

Some time back I watched a talk by Kate Darling about how people respond emotionally to robots. I've been thinking that over as I work with agents. I feel like it's the first technology that can really make me *ANGRY* -- it sometimes feels like the agent is being intentionally dense.

1 week ago 6 0 1 0

Woah man, Firefox Split-View is COOOOOL! Add in vertical tabs, tab groups, and Firefox is really killing it these days.

1 week ago 16 1 0 0

Today's to-do list:

* [x] Argue about AI with people on the internet.
* [ ] Attend governance countil meeting and (I expect) argue about AI.
* [ ] Open up a-mir-formality issues and codify our AI contribution policy.
* [ ] Work on symposium

Heaven help me.

2 weeks ago 45 2 0 0

That too.

2 weeks ago 0 0 0 0

@orhun.dev when we gonna get a ratatui-based Turbo Rust? ;)

2 weeks ago 9 0 2 0

I love these posts :) highlight of my day to see all the awesome stuff people do

2 weeks ago 8 0 0 0
Advertisement

Some days I miss Turbo C++

2 weeks ago 5 1 1 0

Indeed.

2 weeks ago 3 0 1 0

Exactly.

2 weeks ago 1 0 0 0

When I was at my first job, I found a glaring hole in the security of our product. I opened a Bugzilla bug titled "Our security is a joke". I didn't mean anything by it, I was just trying to be funny. But my colleague told me how much it hurt his feelings. I've never forgotten it. Wording matters.

2 weeks ago 85 3 1 2

And how both sides are actually necessary.

2 weeks ago 0 0 0 0

I particularly liked how they exposed the “good cop bad cop” struggle that so often seems to tear allies apart.

2 weeks ago 0 0 1 0

The Gods confound the man who first found out How to distinguish the hours---Who in this place set up a sundial to cut and hack my days so wretchedly into small pieces! . . . I can't even sit down to eat unless the sun gives leave. The town's so full of these confounded dials . . .

— Plautus

2 weeks ago 4 0 0 0
View types redux and abstract fields · baby steps

Bear in mind that I am targeting, to start, only private helper methods. Their impl details are already exposed. If you want to extend to public APIs, I agree you need something more, I covered that here: smallcultfollowing.com/babysteps/bl...

2 weeks ago 1 0 0 0

It knows from the *type*, the same way it knows (e.g.) how many bytes it should be swapping.

2 weeks ago 0 0 1 0

So true.

2 weeks ago 10 0 0 0
Advertisement

Hey, that's my home state! Proud of ya', MD!

2 weeks ago 6 0 0 0

I went to see Suffs last night with my daughter. So many heroes whose stories (and even names!) I did not know. It made me want to learn more.

2 weeks ago 1 0 1 0
Overwrite Trait · baby steps

That would be hard to do-- we would need to have a bound for "swappable" types. And there's no particular reason it can't work. I talked some about the challenges (and potential advantages!) of adding a bound for swappable types in the smallcultfollowing.com/babysteps/se... series.

2 weeks ago 0 0 1 0
Maximally minimal view types, a follow-up · baby steps

New blog post, a short follow-up to the previous one: smallcultfollowing.com/babysteps/bl...

2 weeks ago 9 2 2 0

Interesting thought. Makes sense.

2 weeks ago 1 0 0 0