Just explored NVIDIA's KAI Scheduler (open-sourced by Run:ai (Acquired by NVIDIA) in March 2025) and it directly addresses one of Kubernetes' most frustrating limitations for ML workloads. The problem is simple: K8s traditionally assigns entire GPUs to pods even when your

📣 Join us in Paris for "Foundations for Agentic Workflows" on June 5th @ 19:00! Discover how MLOps must evolve for agentic AI systems ! 📍 Brevo Office, Paris ✨ Limited spots! RSVP now! #MLOps #AgenticAI #ParisAI #TechEvent pourtau marie-laure JennB Register now!

The open deep research pipeline project is coming along. Recruited Alex Strick van Linschoten to ask the real questions (he has a ginger and a black cat) Follow along on GitHub to see progress: github.com/zenml-io/zenml…

this project just keeps growing!

BREAKING: Daytona.io just became the fastest-growing infrastructure company in history. $0 → $1M ARR in 2 months. Faster than Stripe. Faster than Vercel. Faster than AWS. Yes, really. 🧵👇

Infra usually takes 12+ months to reach first real ARR. Why? Because you need a developer to integrate, launch, and then start generating usage. We skipped the infra lag. Daytona hit $1M ARR before most infra companies even get to prod.

Agents or workflows? I asked the community to vote on their understanding. Since the "Building Effective Agents" by Barry Zhang and Erik Schluntz, the industry has been ripe with discussion about what a workflow is vs what an agent is. I was genuinely curious to gauge where we

We are experiencing a Great "Developer" Filter. For those who don't know, The Great Filter is an idea that, in the development of life from the earliest stages of abiogenesis to reaching the highest levels of development on the Kardashev scale, there is a barrier to development

New junior devs who vibe code to learn coding are making devs extinct by 2027

Already tired of the 95% of GenAI projects fail narrative (/re the MIT report that's generating a lot of buzz recently) It's not that we haven't seen this before. In 201,7 we were talking the same number for ML projects in general. Do you know who succeeded in productionalizing

Still in 2025 the easiest way to learn #MLOps: docs.zenml.io/user-guides/st…

Built a lightweight trace viewer to speed up LLM evals—heavily inspired by lessons from Shreya Shankar and Hamel Husain's evals course. Kept it simple: FastAPI + vanilla HTML/JS. Features: failure banner, execution-flow timeline (LLM ↔ tools), keyboard shortcuts, and an annotation

people always ask me, why build custom interfaces for evaluating LLM traces? human evaluation is expensive. custom interfaces make human evaluation 10x-100x cheaper. thanks, Alex, for sharing your example!

Every few years we swear we’ve “solved” AI platforms—then the stack reshuffles itself. What I've seen is a recurring cycle: 🔴 Ad-hoc tools → Scripts, ETL jobs, notebooks. 🔴 Bundling → Opinionated internal platforms (TFX, Michelangelo) nail end-to-end. 🔴 Unbundling → Open