José Montfort

José Montfort

@montfort.dev

Law degree + CS studies. Building StrayMark — governance for AI-assisted engineering. Writing about building it in public.

202 followers479 following236 posts

Pinned

José Montfort
@montfort.dev

StrayMark is a project I've been working on for the past few months. It arose from the need to bring order and purpose to the development of large projects where the fragmentation of human & artificial cognitive capacity made the work truly problematic... StrayMark.dev

straymark.devCognitive discipline for AI-assisted engineering | StrayMarkOpen-source framework + Rust CLI to externalize scope, decisions, and risks so AI agents stay coherent across turns. Maps to ISO 42001, EU AI Act, NIST AI RMF.
José Montfort
@montfort.dev

After months of outright failures with Gemini, Google has announced its first frontier model, Argon 4, but isn't offering it—not even for evaluation—on the Pro plan. They expect me to pay $200 just to give them one more chance. No way.

José Montfort
@montfort.dev

Omarchy pulled out the boundaries that make Linux a system with borders and called the result comfort. On what you learn — and what you'll have to unlearn — when the front door teaches that security is in the way.

montfort.devWhen comfort disarms LinuxOmarchy found the front door Linux had been looking for, and to open it the project pulled out boundaries that took decades of experience to build: agents launched with approvals off, docker group by default, NOPASSWD ALL one command away. On the difference between hiding complexity and removing controls, on what the newcomer actually learns — and on why a project that lights the culture-war fuse cannot later call the smoke somebody else's.
José Montfort
@montfort.dev

This is what I call "the Redot effect"—a fork created by opportunists chasing easy money who exploit far-right political sentiments and possess absolutely no technical relevance. #linux #redot #omarchy

José Montfort
@montfort.dev

DHH needs money and found a way to get it by exploiting the work of those he defames, packaging it under the guise of a distro and advertising it with rhetoric of revenge against an imaginary enemy that oppresses them...

José Montfort
@montfort.dev

I want to try Zed + DeltaDB—give it a shot... But from what I’ve seen so far, it looks like the technical answer to the complaints of "vibe-coders" who don't know how to program, nor how to use Git, nor what it’s even for.

José Montfort
@montfort.dev

And here’s what I mean: I’m sure he’s a good person, but when the change history gets in your way, you aren’t doing engineering—you’re hoisting the sails and letting the wind carry you. You can read the full complaint here, because it says a lot about the problem: x.com/theo/status/...

José Montfort
@montfort.dev

And this makes me think I’m right: Zed’s DeltaDB seems to be the offering for "vibe-coders" who find the discipline of documentation and version control a hindrance—because they aren't doing software engineering; they’re just making requests in a chat. Time will tell...

José Montfort
@montfort.dev

The contract required a notary public to load the reagents and a supervisor for every 150 screens during the application. All the security was placed on the wrong asset. Our opinion on the results of the #UNAM admission exam. strangedays.tech/en/blog/the-...

strangedays.techThe AI didn't cheat: Mexico's online admissions exam — StrangeDaysTechUNAM spent 41.8 million pesos policing its first online exam. It failed. But blaming artificial intelligence is the easiest way to review nothing.
José Montfort
@montfort.dev

I think the main problem with the Codeberg issue is that they are trying to impose a political (anti-AI) position on the projects they host, forgetting that they are just a remote repository and that they do not fund or are part of the governance of them.

José Montfort
@montfort.dev

Torvalds put his foot down on AI in the kernel three times this week, and landed on the same argument I keep making — plus one place where he cuts harder than I do.

montfort.devTorvalds and the mirrorOn a linux-media thread about letting LLMs help kernel maintainers, Linus Torvalds put his foot down three times in favor of AI as a tool. The man whose day job is reviewing other people's patches landed on the same argument I keep making — and cut harder than I do in one place. On where technical authority comes from: the workbench, not the applause.
José Montfort
@montfort.dev

Bun rewritten in Rust: when tribalism drowns the technical debate — Anthropic rewrote Bun from Zig to Rust with 64 instances of Claude, and the community's reaction curdled into tribalism. On technical leadership, self-criticism, and the old…

montfort.devBun rewritten in Rust: when tribalism drowns the technical debateAnthropic rewrote Bun from Zig to Rust with 64 instances of Claude, and the community's reaction curdled into tribalism. On technical leadership, self-criticism, and the old temptation to discredit the dissenter instead of rebutting the argument — and why choosing tools on merit is the opposite of the tribal reflex.
José Montfort
@montfort.dev

It's not atrophy, it's a mismatch — An article making the rounds this week says agentic coding atrophies you. I think 'atrophy' is the wrong diagnosis — and the error isn't innocent: it folds distinct problems into one bag and presents them as…

montfort.devIt's not atrophy, it's a mismatchAn article making the rounds this week says agentic coding atrophies you. I think 'atrophy' is the wrong diagnosis — and the error isn't innocent: it folds distinct problems into one bag and presents them as unsolvable. What actually hurts is the gap between our speed and the machine's — and that gap can be closed.
José Montfort
@montfort.dev

It is said that before becoming an artist manager, Colonel Tom Parker had a dancing chicken show. He would put them on a hot pan to make them jump. Elvis Presley gave his last concert on June 26, 1977; he would die a few weeks later. His life was as brilliant and brief as a sparkler.

José Montfort
@montfort.dev

The immaculate programmer — A reply on Hacker News said the best human programmers would never make the mistakes AI makes. We built an entire discipline — code review, linters, postmortems — on the premise that they do. On the asymmetry of…

montfort.devThe immaculate programmerA reply on Hacker News said the best human programmers would never make the mistakes AI makes. We built an entire discipline — code review, linters, postmortems — on the premise that they do. On the asymmetry of scrutiny, and the programmer we never were.
José Montfort
@montfort.dev

How Mexico City's earthquake alert reached phones in 2012 — A 7.5 earthquake, a brother calmly looking at the sky, and one question: why can't this alert reach a phone? How I ended up running push delivery for half a million people — and what it…

montfort.devHow Mexico City's earthquake alert reached phones in 2012A 7.5 earthquake, a brother calmly looking at the sky, and one question: why can't this alert reach a phone? How I ended up running push delivery for half a million people — and what it taught me.
José Montfort
@montfort.dev

StrayMark is a project I've been working on for the past few months. It arose from the need to bring order and purpose to the development of large projects where the fragmentation of human & artificial cognitive capacity made the work truly problematic... StrayMark.dev

straymark.devCognitive discipline for AI-assisted engineering | StrayMarkOpen-source framework + Rust CLI to externalize scope, decisions, and risks so AI agents stay coherent across turns. Maps to ISO 42001, EU AI Act, NIST AI RMF.
View full thread
José Montfort
@montfort.dev

We also discovered that with the documentation generated on the fly, agents have readily available, well-structured knowledge of follow-ups, pending tasks, and technical debt...

José Montfort
@montfort.dev

StrayMark isn't a prompt, nor does it belong to the emerging category of "harnesses." It's cognitive engineering with a robust framework and CLI. Try it; it's open source and free.

José Montfort
@montfort.dev

StrayMark is a project I've been working on for the past few months. It arose from the need to bring order and purpose to the development of large projects where the fragmentation of human & artificial cognitive capacity made the work truly problematic... StrayMark.dev

straymark.devCognitive discipline for AI-assisted engineering | StrayMarkOpen-source framework + Rust CLI to externalize scope, decisions, and risks so AI agents stay coherent across turns. Maps to ISO 42001, EU AI Act, NIST AI RMF.
José Montfort
@montfort.dev

Currently, AI agents working with StrayMark are generating a real-time knowledge graph of the project based on AIDEC and AILOG documents (among others), discovering, on-their-own, patterns and anti-patterns to improve implementation tasks...

José Montfort
@montfort.dev

Do you know how many workplace conflicts I've witnessed because so many people find it absolutely impossible to admit "I was wrong"? Please don't let AI agents develop egos; what this screenshot shows is a dream.

José Montfort
@montfort.dev

It's been forty years now... I remember the conversations the adults had back then, about the danger of the "radioactive cloud" and when it would reach us. We owe so much to the "liquidators." All honors to them, who saved us from the corruption and incompetence of the ruling classes. #Chernobyl

José Montfort
@montfort.dev

In today's episode of AI Being Irresponsibly Condescending, #ClaudeAI told me: It is the first time I have seen that the complete AI software lifecycle has a coherent, governable, and auditable model. It is no coincidence that it is emerging from real-world use, not from a standardization committee

José Montfort
@montfort.dev

Gemini told me, "Your workflow with your project is a masterclass in Spec-Driven Dev and agent orchestration... your final implementation is, honestly, one of the most robust I've heard lately." Then Claude reviewed his work and told me, "Gemini is dangerously optimistic." What a party pooper 😒

José Montfort
@montfort.dev

Those videotapes waited 50 years to be revealed by a true artist. Thank you, Baz Luhrmann, it was amazing. #elvis #epic #epicfilm #elvisPresley.

José Montfort
@montfort.dev

Speaking of AI taking jobs away from programmers, I keep seeing terribly designed and poorly programmed web apps, redundant workflows, unnecessary data capture, and sections of pages with unprotected personal data publicly visible... No AI would have made such terrible mistakes.

José Montfort
@montfort.dev

#ElvisPresley called #DavidBowie months before his death (1977) to ask him to produce his next album. What a great team they could have made! By the way, it's January 8th, happy birthday to both of them!

José Montfort
@montfort.dev

Well... He's right.

rob pike@robpike.io

Fuck you people. Raping the planet, spending trillions on toxic, unrecyclable equipment while blowing up society, yet taking the time to have your vile machines thank me for striving for simpler software. Just fuck you. Fuck you all. I can't remember the last time I was this angry.

José Montfort
@montfort.dev

Quick update: I'm paying $100 😞

José Montfort@montfort.dev

I'm testing Claude.ai for complex software requirements engineering tasks and it's like a rollercoaster ride: Interactive work and in-depth analysis are like "😱🫣🤯 WOW OMG!!!!" and then I hit the session usage limits like "seriously??? wait for 3 hours or $200??? 😪🙄🫤"