Eleven sessions, one real app you break and fix, and the opinionated calls we made teaching Kubernetes to working developers. Free under CC BY-NC: fork it, run it, teach from it.
Standing up an MCP server is an afternoon's work. Running one that real AI agents hit in production is a different problem. Here's what we got right, what bit us, and how our 91-tool server is actually built.
A CRM is only as good as what your field team puts into it. Here's the approach we've seen work: let reps capture updates straight from WhatsApp and voice notes in the moment, with AI drafting the record and a human confirming before anything commits.
Part 2 of our velocity series. Once one dev-hour stops shipping one point, the things built on that assumption — how you plan, how you hire, how you build capability — break next. And nobody has the playbook yet.
Real orders don't arrive as clean forms. They come through WhatsApp, Telegram, or voice notes — on the apps everyone already has. Here's how we turn them into reliable business objects: AI reasons, MCP acts, and the user always has the final say.
The Model Context Protocol in plain terms: why it's spreading so fast, how it differs from a normal API integration, and the honest answer to when you should use it versus wiring things up yourself. From a team running 195 tools through a single MCP server.
Linear assumes one dev-hour ships one point. With Claude Code, it ships four to seven. Here's how we rebuilt velocity tracking to learn each person's real multiplier from history.







