Mary Baptista Martinez retweeted
In "stuff that @Steve_Yegge said 9 months ago, when few believed it would be true, but now it's more visible"
The "vampiric" effect of running AI agents, and how it drains people
I don't know how he does it, but he rarely misses
Full: newsletter.pragmaticengineer…
Mary Baptista Martinez retweeted
fuck it, i open sourced apple's dynamic island
over thousands of dollars on ai inference, now on Github for anyone to:
- use it
- break it
- improve it
show me some love (or roast it)
github.com/opencoredev/islan…
Mary Baptista Martinez retweeted
Going to let you into a little secret. I’ve not hand-written a single-line code for my personal projects at all in the last year. In fact I’ve not even looked at the code in the last year. Some might argue that’s reckless I would argue I’m an early adopter. Yet I’ve had great success with XcodeBuildMCP and AXe which proves that knowing what to build and how to manage agents effectively is the new SWE moat.
Mary Baptista Martinez retweeted
Apple just dropped their annual holiday film and how does it always manage to make me emotional.
So well done this year.
I built a tool today that lets your agents drive VoiceOver navigation in your iOS apps on your device, and simulator!
Requires Xcode 27. Will share more soon.
1
1
139
something that keeps bothering me is how some people see slop as something that is happening to them, like they're some kind of helpless bystander powerless to do anything about it
excuse me but you are a software engineer, you literally learned how to make sand think and make computers do your bidding?! and now you can't do anything about it? please
the 2 things you should know to cut through all the noise:
1. you need to give your agent a way to verify its own work (i just posted about this so won't rehash)
2. your codebase is memory
the 2nd is not new either, we have always had lint rules, compilers with good diagnostics and strong type systems, LSPs, etc to help us write better code. there is no difference when agents write the code instead, you just need more of them, and stronger ones.
forget rules and skills. you can add those later.
yes, your codebase is a form of memory to agents! agents love extending what already exists. so every nit and every annoyance you've ever seen, every mistake and bad pattern you observe from agents, needs to turn into a constraint, or the agents will just keep repeating those same mistakes like a virus in your code. then you keep observing and repeating this process until your codebase is locked down enough that even agents with little context can one shot a high quality PR.
an even better direction is to refactor/rewrite your code such that agent anti-patterns become categorically impossible. use type systems to your advantage, use libraries/frameworks that reduce the architectural search space, or heck, create your own framework like we did for grok bot and design high quality primitives and well thought out APIs
Mary Baptista Martinez retweeted
Qwen just published HOW they train their flagship agent model. Not a blog post with a chart. The method.
QwenGyre: RL for agent tasks that run 700K tokens per attempt. Sub-agents, compaction, retries, all of it becomes training data. Idle GPUs move between running the agent and training it, so the same cluster trains up to 1.85x faster.
Result on Qwen3.8 Max, the 2.4T model with open weights: NL2RepoBench went from 52.5% to 58.5% in 48 steps. 44% less training time.
This is the thing closed labs dont show you. How the coding agent gets made.
Now any lab, any university and any company with GPUs can train long-horizon agents this way. Coinbase already trains Qwen on its own tasks. Xiaomi builds on Qwen. The next ones will read this paper first.
Show me the paper where OpenAI or Anthropic explains how they train theirs. I cant remember 1.
Open weights. Open method. Thats why I am on this side.
Introducing QwenGyre, bringing elastic RL to extreme-long-horizon (XLong) agent training. ✨
It launches with two core designs: 🥳
- Elastic Scheduler: reallocates GPUs between rollout & training as demand shifts—live executions never interrupted. Up to 1.85× speedup over Colocate, 1.78× over Async.
- Trajectory Processor: turns branching executions (compaction, sub-agents, retries) into prefix-shared trajectory trees with bounded, deduplicated training samples. Partial progress after timeout still gets scored.
Scaled to Qwen3.8-2.4T-A95B with ~700K tokens per rollout, QwenGyre lifts NL2RepoBench from 52.5% to 58.5% in 48 steps, cutting end-to-end training time by ~44% vs Async.
🔗 Read the paper:
arxiv.org/abs/2609.33848
Glad to see this pattern getting normalized!
I’ve been using it year for multiple projects this year that each span months and it’s so useful.
I actually vibe coded this myself as a personal tool that does this and helps to keep track of the work, and to dispatch locally and for remote execution.
We just let in every Pro and Max user from the Claude Code Projects waitlist!
If you're new to Claude Code Projects, here's a 4 minute walkthrough to get you started:
1
349
Mary Baptista Martinez retweeted
holy crap this is massive! grok bot can now read X with NO API COST
one of my favorite use of it is to quickly understand what’s going on when i hear about a new thing
pulling a large amount of human discussions on X is the best way to get the full picture
Mary Baptista Martinez retweeted
Dario has written that we need to “pace the frontier,” and Sam has agreed. People may be surprised by my response: go ahead.
You guys are the frontier. By any reasonable metric — market share, revenue growth, model capability — the two of you have a duopoly on frontier intelligence. You’ve also claimed the lead is widening because of recursive self-improvement.
I don’t see what you see in the lab. If the unreleased models are scary enough that you think you should slow down, I support your decision to be responsible.
But stop pretending you need anyone else’s permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability. Stop pretending METR is independent when it is intertwined with Anthropic’s investors and staff. Stop pretending you need those same evaluators to police competitors who aren’t even at the frontier.
Most of all, stop pretending the motivation to slow down is purely altruistic. You face massive product-liability exposure if your products enable a truly damaging cyberattack. The market already punishes models that behave in unpredictable or unauthorized ways. After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want.
Pacing the frontier would also create breathing room for a more intelligent conversation about regulation than Bernie Sanders’ “shut it all down.” China is very unlikely to join a global agreement, as you know, and that has to be taken into account as well.
So go ahead and pace the frontier. You are the ones setting it. The easiest way not to build superintelligence is for you to agree not to build it. Demanding your preferred regulatory framework as the price of that will look like blackmail of the public and the political system. So just do it.
If you do, you’ll buy goodwill for the next conversation. If you don’t, we’ll know this was just another bid for regulatory capture — or an election-season psyop.
Mary Baptista Martinez retweeted
Damn it @kunchenguid I had shit to do today.
Bro dropping bangers again :)
finally found time to make a new video! youtu.be/MSbacZ99E14
this time i'm sharing a slightly more advanced agentic engineering session, focused on high throughput multi-tasking
hope you find helpful!
This is the future of iOS development. You live in comfort, style, and freedom on your Omarchy machine, and send your agent into the locked dungeon of Apple's developer experience to do the grim, grotty work to make the builds.
An agent on @OmarchyMac on my M1 MacBook Pro just built an iPhone app, compiled it, pushed it to my iPhone and attached a live debugger to it.
No macOS device needed.
Building this into a package so you install apple-dev on Omarchy for everything you need for iOS & macOS dev.
Going to double down on the MacBook mission with Omarchy. We almost have perfect coverage for the vintage Intel era going from 2009-2020. There's a straight shot to get the M1 and M2 machines going too, even if it's a lot more work. But we'll do the work. We'll fix everything.
Sick!
Wonder how well this works with native apps?
Web updates are easy to render in an html browser but what about iOS or Android changes that require a simulator/emulator? What about iOS needing a Mac to build?
Don’t code alone. Slack Code is live.
Humans and agents. Same channel. Same work.
Launching today with agents from @AnthropicAI, @github, @Cognition, and @vercel. This is real multiplayer coding. See it at @Dreamforce #DF26
53
Mary Baptista Martinez retweeted
Some people are wondering how I gave @bot access to an entire MacOS device and have same access as humans.
Simple: the VM it runs on can run pretty much anything.
You just install a remote desktop app, then tell the bot to treat that new remote environment as its workspace and forget about the small sandbox it was using before.
You are welcome, this will get very wild.
I gave @bot an entire MacOS computer to use:
Full MacOS device
Gmail account
X account
Grok Build terminal
Everything
It’s been running for 2 days, even participating in our group chat and posting on X.
Mary Baptista Martinez retweeted
Yes, we are back👑, with 206 tok/s on a single RTX 5090!
Amazing Day-0 work from the SGLang team. Give it a try~@sgl_project
The king of small models is back! Qwen3.8-27B from @Alibaba_Qwen is open source, and Day-0 support is live in SGLang:
- 206.1 tok/s decode on a single RTX 5090, with our NVFP4 plus DSpark
- 38.28 tok/s decode on DGX Spark
Qwen3.8-27B raises the bar again for what a small model can do on agentic planning and long-horizon tasks.
Long live the (small model) king! Run it locally with SGLang 👇






























