Short essays on the small, boring decisions that decide whether a system survives its first six months. Drawn from running an autonomous newsroom, a B2B outbound campaign, a consumer PWA, and a memory layer for one operator.
The what-if audit and the pre-mortem — steps 5 and 6 of the pre-build process. Each catches a category of failure the other misses. Both run in an hour. Most projects ship without either.
A productized outbound stack — sequence engine, enrichment, reply classifier, dashboard — pointed at the audience I knew best. Infrastructure was working. The wedge wasn't. The fix is a fifty-name test that runs before any build.
Of the six steps I run before any code, the whitepaper carries more weight than the other five combined. What it needs to contain to be useful as a contract rather than a wishlist.
A trading bot that printed +18.4% in backtest, +14.7% in mid-price paper-trade, and −3.1% once realistic spreads got modeled honestly. The shelving was deliberate. The pattern transfers to anywhere the cost of being wrong is real.
A six-step pre-build process I run before any new AI-automated system. The cost of an evening's planning is the cost of a quarter's rebuild.