Writing
Essays on software engineering — building well, running lean, and making the most of what AI has changed. By Matt Tanner.
'AI-native' sounds like an architecture but says little about its quality. Putting a model at the centre of a business system is the wrong goal. Build good software that AI can use, and that stays trustworthy when the AI gets it wrong.
→ 17 September 2026AI coding agents can do real work. They cannot bear consequences, and most production systems were built assuming that whoever wrote the code could. The question that matters now is not how the code was produced but what it is allowed to do.
→ 20 August 2026Long, open-ended consulting relationships often produce the opposite of what clients need. A focused engagement — days or weeks, not months — forces clarity on both sides and produces better outcomes.
→ 16 July 2026A specification describes a solution someone has already imagined. The more useful question is what the system actually needs to do — and that question is often never asked.
→ 18 June 2026Most of a system is fine. The trouble concentrates in a handful of places where the structural decisions actually matter — and that is where attention belongs.
→ 14 May 2026The full rewrite is the most expensive decision a team can make, and usually the least examined. Most of the time the system is telling you something more precise.
→ 16 April 2026The patterns reproduced by AI tools are only as good as the code they were trained on — and most of that code contains accumulated compromises. Starting from the intrinsic nature of a problem is harder, but it's the only reliable path to a system that fits.
→ 12 March 2026AI tools have made it faster and cheaper to produce code than at any point in the history of the industry. The constraint has moved — and most software processes haven't caught up yet.
→