Sagir Khan

Agentic AI @AWS · Open Source @OpenJSF

[Sagir Khan @sagir_k]

1/ I used Codex loops for a month in brownfield production systems: ~$100k of inference, ~20.4B token-throughput, 393 sessions.

Real customers, auth-gated workflows, flaky local envs, careful maintainers, years of accumulated decisions.

🧵 A few field notes.

[Kenton Varda @KentonVarda]

Building software by prompting agents without ever reading or editing the code directly isn't actually a new thing. It's just called being an engineering *manager*. We have decades of experience about how to be effective as engineering managers. It is a different skill, though.

[Jensen Huang @JensenHuang]

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.

AI will transform every industry, power every company, and be built by every country.

Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

The world needs both frontier closed models and frontier open models.

https://t.co/AUKzoQ5Ikb

[Marc Brooker @MarcJBrooker]

Our new paper on the design and implementation of Aurora DSQL is now up. I'm super excited about this: it brings together end-to-end view of how DSQL works, where we're going, and what we've learned.

I've also put together a fun interactive transaction demo.

[Felix Rieseberg @felixrieseberg]

Have you built a language model? You should. It's so much fun to chat with something you made.

Anyone can do it, too. I made an app that teaches the fundamentals and gives you everything you need to build your own: https://t.co/nLjFfpczbV

[almonk @almonk]

https://t.co/Rz0q1W33ut

[Mitchell Hashimoto @mitchellh]

https://t.co/RZlp39Wkrf

[dex @dexhorthy]

its almost as if I wrote an entire AI Engineering guide who's core thesis was on how graphs are better than loops

over a year ago

enjoy - https://t.co/9iXhtq1qvd

[Brace @BraceSproul]

https://t.co/GUcQHOYf6H

[Mario Zechner @badlogicgames]

recommended reading. simple is good, simple is beautiful. simple works.

[Glauber Costa @glcst]

I am excited to announce that we are officially writing a new version of Postgres. In Rust - and creating the LLVM of databases in the process.

In the span of a year, we have rewritten SQLite. Keeping the compatibility, increasing its feature set. MVCC, Types, (Live) Materialized Views, among other things. In the process of doing that, we have realized: At the end of the day, what makes SQLite special is that it compiles SQL to a database-specific bytecode. So why can't we compile *Postgres* to the same bytecode?

Turns out we can. I ran an experiment called pgmicro as a way to prove this hypothesis, and it works very well. It is time to make this official, and put the weight of Turso behind it. We shall give the world a modern take on Postgres. Wire compatible, but built on a new architecture.

We have already heard of others wanting to extend this. MySQL? Redis? the sky is the limit. What can we do if we do for databases what LLVM did for compilers? To prove how powerful the SQLite bytecode is, we are actually running DOOM compiled to the unmodified SQLite instruction set. And because Turso runs natively in the browser, you can play the game in your browser. With the database executing it.

Read the full story below! 👇

[Peter Steinberger 🦞 @steipete]

OpenAI hired me, not OpenClaw.
The OpenClaw Foundation is independent, with sponsors rather than owners - and, for the first time, a full-time team keeping the claw alive and stable. 🦞

Couldn’t have done it without Dave and the team. Kudos!

[Anders Hejlsberg @ahejlsberg]

Huge milestone for our team today: TypeScript 7 is now generally available--a native port that runs 10x faster. @typescript https://t.co/3DtJJqEZAl

[Jarred Sumner @jarredsumner]

Rewriting Bun in Rust

https://t.co/Rl8bcaxBFc

[Dave Morin 🦞 @davemorin]

We may have finally created the bicycle for the mind.

[Maxime Rivest 🧙‍♂️🦙🐧 @MaximeRivest]

Fable turned my remarkable into Tom Riddle's diary from Harry Potter.

My prompts fade, a LLM respond.

Magical!

[Sagir Khan @sagir_k]

@FardeemM This is excellent because it’s executable advice.

I pointed Codex at it during a brownfield React rewrite and it turned into gates: generated clients from backend contracts, one server-state path, typed routes/search, a component system.

“Build the factory” is exactly right.

[Sagir Khan @sagir_k]

@steipete @nateschnell @bcherny I have a much smaller version of this running on brownfield work, and the wild part is the control plane adjusting itself on the fly.

Goals, heartbeat, active lanes, blockers: each pass reshapes the next one.

Once you see that, it’s hard to go back.

https://t.co/WD2p4YWhHS

[Sagir Khan @sagir_k]

9/ One pattern I didn’t expect: control-plane loops.

Some threads should not code at all. Their job is to track lanes, avoid duplicate work, preserve context, collect proof, and decide the next safe handoff.

[Sagir Khan @sagir_k]

@shadcn The brownfield version is brutal:

you ask for one change, the loop finds five adjacent almost-problems, and suddenly the patch is “cleaner” but no longer reviewable.

A good loop has to notice nearby work without absorbing all of it.

https://t.co/fpplVJsMMj

[Sagir Khan @sagir_k]

6/ Brownfield software is full of almost-relevant work:

old abstractions, naming drift, half-finished migrations, test gaps, tempting cleanup.

A good loop notices those things without absorbing all of them.

[Sagir Khan @sagir_k]

@ScriptedAlchemy Same. Huge swarms are magical, but for real work I keep coming back to a few visible lanes.

Sidebar threads are finite, inspectable, resumable, and easy to drop into side-by-side.

Makes intervention convenient and cheap.

https://t.co/WD2p4YWhHS

[Sagir Khan @sagir_k]

9/ One pattern I didn’t expect: control-plane loops.

Some threads should not code at all. Their job is to track lanes, avoid duplicate work, preserve context, collect proof, and decide the next safe handoff.

[Sagir Khan @sagir_k]

13/ The thing I’m most excited about is not agents running forever.

It’s agents becoming useful in brownfield software: existing systems with owners, users, tests, migrations, and review norms.

That sounds modest. But in work people already depend on, it compounds fast.

[Sagir Khan @sagir_k]

1/ I used Codex loops for a month in brownfield production systems: ~$100k of inference, ~20.4B token-throughput, 393 sessions.

Real customers, auth-gated workflows, flaky local envs, careful maintainers, years of accumulated decisions.

🧵 A few field notes.

[Sagir Khan @sagir_k]

12/ My current rule of thumb:

If I can’t review the loop result quickly, the loop was too wide.

Shrink the unit of work until it feels like a clean PR, not an archaeology report.