Writing
Startups, training, and the small experiments in between. Roughly one post whenever something turns out to be worth writing down.
2026
11 posts01Making Composio production-readyOne integration puts hundreds of tools in reach. Here is everything we had to build around it before it was safe to point at a real inbox.02What actually moves time to first tokenTwo months of latency work on a production chat agent: the wins, the ideas the measurements killed, and the numbers behind both.03The tool definitions were the slow partA 100% cache hit makes tokens cheaper, not invisible. Deferring the tool contracts halved our chat's time to first token.04Four ways my evals lied to meEvery harness bug masquerades as a model result: false reds read as regressions, false greens read as coverage.05Make your agent review its own work before you pushFanning out parallel reviewers catches confident wrong answers. It also produces confident wrong findings.06Taking the keys out of RillReplacing a plaintext MCP secret with OAuth, publishing the SDK, and moving a live site off Sanity.07Zone 2 Training: Low, Mid, or High — Does It Matter?08Startups Are Not a Marathon—They’re Marathon Training09The only 7 knots you will ever need10How zone 2 changes your muscles11How urge surfing helps resist unwanted impulses