The story of building Aroi Care — a bilingual EN/TH AI food-safety advisor — from a one-line idea to a live app with a Gemini-powered analysis pipeline, two-tier caching, and a hard budget kill switch.
A Technology Polymath who has spent 20 years telling clients to embrace AI finally turns the mirror on himself. What followed was a zero-server, near-zero-cost blog built almost entirely with Claude — and only mildly embarrassing.
Someone asks how long it will take. You say six weeks, because saying anything else feels evasive. Six weeks then becomes a promise you never actually made, and the trouble starts there.
An AI trading agent can search a million variations of a strategy in an afternoon. That is exactly the problem. When the search is cheap, the beautiful equity curve stops being evidence and starts being an accident you paid for.
Requirements describe what someone wants. Constraints describe the shape of the world they are actually operating in. Learning to ask for the second one is the fastest way to stop building the wrong thing beautifully.
Most teams shipping AI features cannot answer the simplest question about them — is this actually good? Evaluation is not a testing chore. It is the definition of correct, and writing it down is the hardest and most valuable work in the project.
Most requests for a decision arrive as a meeting invite or a wall of context. A one-page brief gets you an answer faster, and quietly improves how people see your judgement.
Most organisations have given their AI agents real power and borrowed credentials. The identity question sounds like plumbing, but it decides whether an agent is auditable, revocable, and safe to expand.
