[ WORDS — 8 ARTICLES ]
Writing on engineering, AI & product
- 01Human in the loop is not a buttonMost "human in the loop" designs are a confirmation dialog bolted on after the write. Here is what it takes to build approval into an agent properly: gates, evidence, reversibility and the slow, human parts.
- 02MCP vs in-product AI: where should a job live?When a customer wants an AI to work with your product's data, should the job be served by an MCP server in the assistant they already use, or by AI inside your product? A working test, and why the line probably runs along consequence, not surface.
- 03What I'd tell the next PM on an AI teamLessons from leading an agentic chat product inside an accounting platform: learn faster than you build, put legal and safety inside the harness, back your engineers' experiments, and be strict about the evidence even when you're loose about the code.
- 04Stop wrapping your API: designing MCP tools around outcomesThe first version of most MCP servers is a thin wrapper over an API. Here is why consolidating into a handful of outcome-shaped tools makes the model more accurate, the consent screen honest, and the whole thing easier to gate, audit and test.
- 05Your eval has one accuracy number, and it's lying to youA single end-to-end accuracy score hides whether your agent picked the wrong action or the tool did the wrong thing. Those are different bugs with different owners, and you need to measure them separately.
- 06The sub-agent bubble is quietly deflatingSub-agents were a clever solution to a temporary constraint. The models got better. A different pattern is winning: one that bakes capability directly into the experience instead of routing it somewhere else.
- 07Stride MCP: when your MLP hits the athlete cap before anyone runs
- 08I Built an End-to-End AI Running Coach in One DayHow I used Claude Code and Cursor to ship a full-stack AI coaching app in a single day: durable workflows, LLMs grounded in real data, and the boring middle of indie engineering.