https://abdulbasitkhan.com/ Essays and field notes by Abdul Basit Khan en https://abdulbasitkhan.com/notes/decision-records-for-the-inheritor/Code shows what a system does. It rarely shows what you deliberately chose against, or why. A few lines of decision record saves the next person from re-solving a problem you already solved.Mon, 20 Jul 2026 00:00:00 GMThttps://abdulbasitkhan.com/notes/decision-records-for-the-inheritor/https://abdulbasitkhan.com/essays/one-rule-file-thin-adapters/Every code editor now ships its own AI agent, and each wants its own config file. Duplicating your rules per tool is how they drift apart. Keep one canonical rule file and make everything else a pointer.Thu, 09 Jul 2026 00:00:00 GMThttps://abdulbasitkhan.com/essays/one-rule-file-thin-adapters/https://abdulbasitkhan.com/notes/local-inference-is-economics/Running models on your own hardware is usually sold as a privacy move. The more honest reason, most of the time, is cost and latency control during evaluation.Wed, 08 Jul 2026 00:00:00 GMThttps://abdulbasitkhan.com/notes/local-inference-is-economics/https://abdulbasitkhan.com/notes/provider-adapters-cheap-insurance/Wiring your app directly to one model provider feels fast. A thin adapter behind a stable interface costs a little now and saves a lot the day pricing, quality, or availability changes.Wed, 24 Jun 2026 00:00:00 GMThttps://abdulbasitkhan.com/notes/provider-adapters-cheap-insurance/https://abdulbasitkhan.com/essays/intent-is-not-reality/Most project docs rot because they mix what a team decided to do with what is actually true now. Separating intent from verified reality is the fix, and it matters more once agents read your docs too.Thu, 18 Jun 2026 00:00:00 GMThttps://abdulbasitkhan.com/essays/intent-is-not-reality/https://abdulbasitkhan.com/notes/kill-switch-is-a-feature/If you cannot turn an AI off in one move, you do not really have it under control. Treat the off switch as a product requirement, not an ops afterthought.Wed, 03 Jun 2026 00:00:00 GMThttps://abdulbasitkhan.com/notes/kill-switch-is-a-feature/https://abdulbasitkhan.com/notes/confidence-across-languages/A model's confidence is not a fixed currency. The same number can mean different things across languages, so a single global threshold quietly under-protects your second language.Wed, 20 May 2026 00:00:00 GMThttps://abdulbasitkhan.com/notes/confidence-across-languages/https://abdulbasitkhan.com/essays/confidence-is-not-permission/Most teams let an AI act when the model feels sure. That is the wrong gate. Here is why confidence and permission are different questions, and how an autonomy ladder keeps AI useful in production.Tue, 12 May 2026 00:00:00 GMThttps://abdulbasitkhan.com/essays/confidence-is-not-permission/