Giustino 🍩 retweeted
If by "Software Factory" you mean another grandiose name for a collection of scripts, loops, and workflows, invented by people who need something new to put in their pitch decks, a phrase designed to get developers excited and VCs to open their wallets, a bandwagon for people who have discovered cron jobs and would now like to raise a Series A, then I am against Software Factories.
But if by "Software Factory" you mean a real and rather interesting change in how software gets built, where engineers who were previously using agents to write features and fix bugs are now building systems that get agents to do those things autonomously, where the engineer has moved up a layer of abstraction, from working on the product to working on the machinery that builds the product, then I am for Software Factories.
Because that's the distinction I'm interested in. We're no longer just using agents to build software. We're building the thing that uses agents to build software.
You can call them loops. You can call them workflows. You can call them whatever you like. But "Software Factory" seems as good a name as any, and I rather like the idea that we're no longer merely product engineers, but factory engineers.
This is my stand. I will not compromise.
Just wanted to defend the term "Software Factory" for a second.
First, the caveats:
It's bullshit. It sounds fancier than it really is.
It's a bandwagon. People are using it to get you and their VC's hyped.
But it's also referring to a real change.
Earlier this year, most SOTA teams were using agents to work on the product. Shipping features, fixing bugs.
But now, they're building automations to help agents work on the product for them. They've moved up a layer.
What do we call the thing they're building? Loops? Workflows? Some other generic label?
"Software Factory" is as good a term as any other.
- "We're not working on the product, we're working on the factory that builds the product."
- "We're not product engineers, we're factory engineers."
I like it. I'll continue using it.
turns out it classifies what audio it hears, before reporting back to samsung.
sensor preferences, dog-barking and baby-crying event counters, and camera streaming duration are built into telemetry paths.
Giustino 🍩 retweeted
We’ll still need end to end tests, and frankly, we’re going to need all the forms of testing I’ve been talking about for a decade.
But “unit tests are now slop” doesn’t have to be the foregone conclusion.
Unit tests can be very useful even if a human will never again read them.
We in Europe have a lot of problems in creating companies. But a big one was getting funding to pay programmers and create big companies. I believe that with AI, if people change mentality a little bit, even just the young folks, we can move a lot of IT back to Europe.
Giustino 🍩 retweeted
This analogy is the best I've ever heard when it comes to how LLM-generated code looks, works, and what the role of developers is morphing into.
Pre-AI codebases are sturdy, with scaffolding that allows for human comprehension and, at the same time, prevents hyper-optimization. We need to "get" the code, not just run it.
Post-AI codebases let go of all pretense to be both accessible by interpreters/compilers/runtimes AND humans, and contain JUST what is needed to provide the expected outcomes.
And that is what developers are realizing they must measure, review, and verify. Not the code, but the constraints and tests around it.
Codebases are going to start looking like optimized topology 3d prints
When AI is better at math than humans it also becomes better at algorithm & codebase design
At that point the system and delivering quality software doesn’t require human understanding to exist in the actual code layer.
Only needs to be understood by humans at the high level of passing testing, verification, vibe checks and being understandable when queried via AI in the rare case one need to dive deeper.
Giustino 🍩 retweeted
As reported in the @WSJ, we have raised a Series B to bring Hermes Agent to new frontiers (and, yes, build a mobile app)
Thank you to our investors including @nvidia @M12vc @SamsungNext @robotventures @usv @ycombinator @MenloVentures, among others
wsj.com/pro/venture-capital/…
sharing my thoughts on our recent fundraise.
we did alright for a couple of goofballs
nousresearch.com/a-note-on-o…
Giustino 🍩 retweeted
Today, we are launching a preview of our new model, Mistral Large 4 (ML4), aka le Chonk 🐈.
ML4 is a 1T-parameter model with 49B active parameters, trained natively with multimodal capabilities.
It is at the frontier of open models, and by far the strongest open-weight model from the US or Europe.
The RL run behind this preview is still in flight and shows no sign of saturation -- we will release a final version before the end of the month along with the weights of the model.
🧵
1/n
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
Giustino 🍩 retweeted
Meet Mistral Large 4, aka Le Chonk.
• 1T parameters, natively multimodal. 49B active.
It is the best open weights model from US or Europe on aggregated benchmarks.
• State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding.
• Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure.
• Available to all via API today. Working with cybersecurity partners privately.
Open weights release end of October.
🤖 Made with AI
Giustino 🍩 retweeted
Yup.
Sounds like everybody worked in a Workshop with some real Code Craftspeople that hand-chiseled code every day.
On pacemaker software, avionics software, and, of course, the code that keeps nuclear arsenals safe.
Replying to @thorstenball @badlogicgames
Nope and I think pretending like so much handwritten code was artisanal and not half copy pasted stack overflow edited until it runs / is compiled is silly.
Giustino 🍩 retweeted
Karpathy: disappears from X
Ben Affleck: alright, gather round, so you'll want to freeze the base weights first, learning rate 2e-4
Ben Affleck (Hollywood star & Artists Equity CEO) talks about how he fine-tunes open video models by unfreezing weights and trained only the last cinematic layer so a film crew can hit real production standards.
for context, Ben Affleck founded InterPositive in 2022, a 16-person AI shop for film post and Netflix bought it in March 2026 for $587 mn in cash.
He needed that model because public video models were trained on his peers' films, and he did not think that was a real business. So InterPositive raised money, shot its own dataset for 8 months on a controlled stage, and used it only as late-stage training.
Each new film then trains a private model on its own dailies, so the production keeps the footage and the learning. That is the product Netflix paid $587 million for.
----
From "Bloomberg Live" YouTube channel, (link in comment)
Giustino 🍩 retweeted
Experience gives perspective. Programming is not dead, and never will be.
It's late September 2026. Sometime in September 1976, just after I entered college, I wrote and ran my first serious program. I have now been a programmer for literally 50 years.
And man, I have seen it *all*. Punched cards to glass ttys on minicomputers to workstations to the earliest micros. I've written business software, communication programs, development tools, games, language implementations, kernel device drivers, graphics libraries, a few critical pieces of infrastructure, and all manner of odd programs for which there is no taxonomy. Coded in a score of languages some of which are now forgotten. I've run over 70 open-source projects and contributed to at least as many more.
Along the way I noticed a few things, and wrote about them, and that made me famous. But I'm not here to write about that today. I'm here to write about what it means to be a programmer and a technologist with the Singularity upon us.
I have lost count of the number of times I have been told "programming is over". You kids may think this is a new claim that came into the world with LLMs. Hah! There's been that kind of talk at every step function in the quality of our programming tools. I first heard it when compilers were replacing hand-hacked assembler in the early 1980s, and I wouldn't be a bit surprised if somebody said it earlier when in-memory code replaced plugboards.
One constant amidst all this change is that the intentions in a user's head do not magically turn into executable instructions that a computer can run. Another constant is the mindset of the people who specialize in bridging the gap between wish and algorithm. A programmer is a programmer is a programmer - the tools are secondary.
"But it's different this time!" some people will say, "AIs are superhuman now!" So? What did you think the optimization stage of a compiler was, chopped liver? We've been relying on tools that exceeded human capabilities all along - we built them precisely because our meatbrains could not otherwise keep up with the demands of the job.
It's been a hell of a ride, that 50 years. I've succeeded and failed and succeeded again. Done a lot of work I'm proud of. Built at least a few things that look like they're going to last. None of that goes away because, hypothetically, an AI might have made me faster.
The work is never done, even if the tools can talk to us now. Because intentions still have to be communicated; people who specialize in that and develop right mindset to do it still have a major advantage over people who don't.
If you're feeling depressed because you spent years or decades developing hand-coding skills that are obsolete now? Buck up. You still have the most important asset - a mind that knows how to think that way.
I'm facing the winds of the Singularity with a smile. Because 50 years and I'm still standing - still shipping code that people can use, still learning, still refusing to be paralyzed by the pace of change.
The best therapy for your uncertainty? Get to work on something.
Giustino 🍩 retweeted
It’s as if we had robots playing in the NBA along humans yet people were like:
“But if they just mechanically reproduce shooting motions they learned during training*, don’t sweat, and can’t do that crossover 99.9% of people can’t do either, are they truly playing basketball?”
(*: as if it weren’t similar enough to how basketball players learn to shoot)
Good programmers state they use models like GPT 6 Astra without obtaining good results: it makes me feel I live in a parallel universe. But the explanation is not that hard: good programming in the past and now requires a different skill set, even if there is some overlap.
Giustino 🍩 retweeted
Sorry, this is pure FUD. Google API keys are very simple, people just don't follow very simple steps.
All you need to do is:
1. Sign in to google cloud console
2. Create a new project
3. Enable billing on that project
4. Search for numerous "services" that you need to enable for that API
5. Find out there are more you need to enable to actually make it work
6. Go and enable them.
7. Create the key
8. It still doesn't work and you keep wondering why and then spending days with customer support (after many different wrong AI generated answers that lead you to a wild goose chase), you find some hidden webpage that oh wait, you need to create a key in ai studio for this now
9. ai studio says you are only on tier -43 so you can only do transactions of less than $0.2 in a day
10. Then you go to openrouter and make a key and use it.
It's extremely simple.
Giustino 🍩 retweeted
Microservices are what happens when your organization has communication problems and you decide the architecture should have them too.
Giustino 🍩 retweeted
Agent: *screenshots the page*
Agent: *scrolls*
Agent: *screenshots again*
Agent: *clicks the wrong button*
Agent: *screenshots again*
Page with WebMCP: hi, here are the 4 things you can do here, with descriptions
Agent: oh thank god








































