Remote MCP documentation just dropped ✨ ⚡️ Build, test, and deploy remote MCP servers on Cloudflare in minutes ✅ OAuth 2.1 automatically included (+BYO provider) 🧪 Use mcp-remote proxy to test e2e with MCP Clients (Claude, Cursor, et al) developers.cloudflare.com/agents/guides/…

Prediction: in the next few months, Deepseek will release an “emotional” model to compete with 4.5 and a native image generation MoE

SF I am inside you.

Anyone vibecoding a Europa war game?

i wouldn't announce i'd won the lottery, but there would be signs

We started 8090 Solutions Inc. exactly a year ago. I’ll be sending a more detailed update in a few days that explains what we’re up to, but in a nutshell, we’ve been building a highly opinionated software factory that makes extremely reliable, hi value production software for

If you or anyone you know is working on something like this, DM me.

I’ve said it before, but I really think Cluely is going to win. Writing off a product because their marketing strategy offends your sensibilities is stupid.

Full-precision, undistilled, SOTA Models on Groq will change everything

If you think the 100M TC is a lot wait until Meta starts going after the xAI waifu engineers

Very excited about what we’re building. More details to come. If you’d like to try Software Factory in September join the waitlist at 8090.ai/waitlist

What I would build if I wasn't building Software Factory @ 8090: Batteries-included fork of Roo Code with codebase indexing (turbopuffer ), fast apply model (Morph (4,500+ tok/sec)), and Kimi-K2 on Groq Inc for inference. TUI-style user interface for the chat. Call it Fast Code or

I request about ~1M tokens a day from Claude 4 Sonnet using Cursor. I'd say this is about my max. Even if the context windows get larger and the time horizon gets longer, is it going to matter to me? If I can't review it I can't ship it. Even if the model doesn't make any