jul 17, 2026kimi.com ↗
Kimi K3 Brings the Open-Model Gap Down to Weeksmy take: K3 does not lead the frontier, but it shows that open models can now reach it within weeks.
Hi — I'm Stan. Welcome in.
I'm a software engineer and a YC founder, building AI-powered tools for how people work. This is where I write essays on architecture, AI engineering, and product building, share what I'm reading, and leave tools out on the bench for anyone to use. Stay a while.
follow via rss →or find me on github ·twitter ·mastodon
writing/essays, monthly-ish
★ latest · 2026-07-14 · 8 min readYour Agent Has Too Many ToolsMCP tools consume context and affect prompt caching. CLIs and tool search point toward a better pattern: progressive disclosure for agent capabilities.2026-07-05The Great Token Arbitrage2026-06-24User Experience Is No Longer Just for Users2026-06-19Beyond RAG: Agentic Context Constructionall 16 essaysbrowse →
reading/links + my take, weekly
jul 17, 2026vercel.com ↗
Kimi K3 Quietly Arrivesmy take: the benchmarks will decide how capable Kimi K3 is, but its cache-friendly dynamic tool loading is already a meaningful step for open-model agents.
jun 25, 2026snscratchpad.com ↗
A Frontier Ecosystem for Human Learning Loopsmy take: own your learning loop or cede the value to a few frontier models — AI should amplify human capital, not extract it.
2–3 links a weekfull log →
▚ Workbench
all tools →Free tools and playgrounds, born from real problems. Use them here, or embed any of them on your own site.
every tool is embeddable — one iframe snippet, free, no tracking.browse embeds →