Rafal Wilinski
@rafalwilinski
AI Agents Lead @ Zapier
Can you build a full-stack AI app on @cloudflare? The answer is: absolutely. 🔥 Introducing "Fullstack Cloudflare RAG" aka Chat with PDFs solution built 100% on Cloudflare using only their services. Link to demo 👇
Big unlock for CodeAct and other agentic architectures that leverage running LLM-generated code. Huge!
We're excited to introduce Cloudflare Sandboxes — the easiest way to run LLM generated code on demand. Sandboxes make it easy to spin up new instances of containers (and soon Workers — stay tuned!!) to make it easy to spin up tens, hundreds, or thousands of code-running agents
Big news for European AI startups. Complying with regulations will be much easier
Free weekend means I had too much time to play with newest @cloudflare primitives. This is going to be an ideal platform in the future, but for now: - wrangler.jsonc is better than toml but they should just buy Alchemy or SST - containers are great promise but UX and visibility…
Friday night activity - trying to run Cloudflare containers but no success or logs so far.
my dream is to be able to all that RL, GRPO, SFT et al shit in Typescript someday
Our #AIEWF 2025 talk about evals is finally live youtu.be/blrovBxxN9o?si…
Y’all really use this?? Using GitHub Mobile is infinitely better
demystifying hot AI terms: agentic rag - retrieve a few of times sft - few shot prompting but harder synthetic data - source? I made it up evals - tests repeated 10 times
Cloudflare and Vercel release Code Sandboxing service on the same day. I wouldn't like to be a startup in this space right now 💀
True Serverless containers-as-a-service with scaling to zero is finally here.
Cloudflare Containers are now available in public beta. Deploy simple, global, and programmable containers alongside your Workers: cfl.re/4kNjs06
HYPE HYPE HYPE
📦 🔜 Join @MikeNomitch_CF and friends at @coderhq @ValDotTown live online tomorrow (June 24) for updates, demos, and AMA. 🔗containers.cloudflare.com
if your company is not encouraging more token spend by devs you're ngmi
stories from trenches: something that other team estimated as 3 weeks of work was shipped in 1-hour long Claude Code session