Notes & essays
On writing, building, and the spaces in between. New posts land here first.
I don't have a story about watching retrofit accessibility fail; I have one about the accessible thing that never got built. Overlays are wrong, accessibility has to be baked in, and in the time and tokens era building it in should be well within the realm of possible. Today's tools are the floor, and the people who use them every day will judge what comes next.
Part 1 said a capable person who understands the business can now direct AI tools to build working software, and never said which person. The answer has little to do with product versus engineering and almost everything to do with whether you can tell when the machine is lying to you. The gate that used to read 'can you write code' has moved - the mistake now is assuming it disappeared.
AI compressed a six-month, two-million-dollar project into two weeks, and every old way of pricing it now screams at somebody. T&M is dead twice over - the escape is outcome-based pricing done with discipline, so you finally get paid for judgment.
Buying software and bending your process to fit it was the right answer for decades. AI collapsed the cost of building - but the team that keeps bespoke software running still costs what it always did.
The three reasons I structured Gairm as a nonprofit - stakeholders, transparency, and trust - and the one thread underneath all three: servant leadership.