Predictable Machines

We just open-sourced lean4-markdown: generate Markdown from Lean 4 with a typed AST instead of string templates. First of several libraries we're publishing as we build Predictable Code ❤️ predictablemachines.com/blog/lean4-mar… @leanprover #OpenSource

Worth reading if your team is shipping AI-generated code and thinking seriously about what correctness guarantees you actually have. predictablemachines.com/blog/ai-thinks…

None of these are exotic. The point is that they are properties of the whole system, not of any individual function, and that is precisely where testing and LLM review stop being reliable tools. You can cover every line and still not be verifying the thing that matters.

Fran from our team published something I have wanted us to write for a while. Not a post about AI being good or bad at code review, but about what kind of correctness claim you are actually making when you say "the tests pass" versus "this has been formally verified."

What makes a verification system trustworthy? As AI accelerates software creation, trust becomes the real bottleneck. More on how we think about building transparent verification ↓ predictablemachines.com/blog/the-funda…

Predictable Machines is officially headquartered in Seattle's Ballard neighbourhood. Two offices, two coasts, one mission: AI that industry can actually trust. Stop by to say hi at any of our offices in Cádiz or Seattle!

I'm lucky to be part of a founding team with people in tech for the second time in my life. 15 years later, the important parts are still the same. But the context has changed, and it says a lot about the times. I wrote about what it's like being part of @aipredictable : jorgegalindo.me/en/blog/posts/…

AI-generated code moves fast, but consistency is the real challenge. We work at the intersection of software and mathematics, using @leanprover to formally verify that AI-generated code stays aligned with project requirements. Why we chose Lean: predictablemachines.com/blog/why-we-ch…

AI assistants like Claude Code, Cursor, and Copilot write more code every day. The real problem isn’t generation. It’s drift. Code that looks right but slowly moves away from intent and rules. Why this matters (and what we’re building to fix it) 👇 predictablemachines.com/blog/why-we-ar…

Today, the Cádiz office is full of life, good food, and great conversations. But what really matters is being surrounded by people aligned with what we’re building at Predictable Machines. We’ll do amazing things together ❤️

Here, you can see our very own @raulraja explaining one of the products we are working on and validating the product-market fit of Predictable Code. Lack of predictability in AI-generated code is definitely a pain

Lunch might be ongoing but for our sponsors @jetbrains and @AIPredictable its a great chance for speaking the attendees about what they offer and also any potential jobs on offer #sponsors #facetoface #lambdaworld #AI #Kotlin #Ktor #jobsearch

And @Lambda_World folks, if you wanna know more about our tools in the field of Predictable AI, come by our booth! We’re excited to meet you. BTW, we’re hiring kind and smart people! 🚀

Hey @Lambda_World attendees! We have some Predictable Chicharrones (de Chiclana) waiting for you at the lunch area!

Proud sponsors with @AIPredictable ❤️

🚀 We’re proud to support @Lambda_World in Cádiz as Platinum Sponsor! At Predictable Machines, we’re building the future of verifiable & reliable AI, and we’re hiring. If you’re curious about what we’ve got in the oven 🔥, come talk to us at Lambda World!

¡Vamos allá otra vez! Ilusionado como la primera vez. Hapiness is only real when shared. ❤️