What I learn building React apps, Node.js APIs and AI agents for my clients. Guides, hands-on insights and architecture decisions. And sometimes, thoughts on what it means to be a developer in the age of AI.
The first class of System One models just shipped. How the math changes when a typed decision costs 70 milliseconds instead of two seconds, then three places where it shows.
Nothing to install, no landing page, no empty state. What MCP changes about SaaS acquisition and activation, plus the five levers that cost a week.
Fifty tools are resent in full on every call, and the model picks worse for it. The catalog pattern in plain TypeScript, then the Claude API tool search.
Splitting an agent into specialists breaks four very specific things in production. The principle that makes them impossible: one writes, the others think.
The cost of iteration has dropped, the cost of a false signal has not. Three traps, five principles and twelve questions to answer before you code.
MCP connects agents to tools, A2A connects them to each other. Agent Cards, task lifecycle, the three-layer stack, and what production already demands.
Server Adapters, Composite Storage, Observational Memory: what the stable v1 changes for your TypeScript agents and when it replaces three libraries.
The four patterns that make an agent run, four frameworks compared, then integration into an existing project: isolation, testing, scorers, production.