Skip to content
← Journal index
Tools28 April 20268 min read

What I mean by AI-first

AI-first isn't 'wired-up ChatGPT in a sidebar'. It's a different shape of working day, and a different shape of invoice for the client at the end of it.

There are two ways to mean 'AI-first'. The first is the marketing one: a vendor stuck Claude in a wrapper, gave it a brand name, and now charges £40 a seat a month. The second is the studio one: every step of the build, from spec to ship, is shaped around a senior engineer collaborating with a capable agent in the loop. I mean the second one.

In practice that means: schemas drafted in plain English and refined into Postgres in 20 minutes; component libraries scaffolded by Claude and then hand-tuned by me for taste and accessibility; copy iterated against a brand voice doc that the model reads on every pass; CI checks that include an LLM reviewer flagging things human reviewers tend to miss on tired Fridays.

The point isn't that the AI does the work. It's that the AI removes the boring half of every task, leaving me free to spend the hour on the half that matters. Design decisions, performance tuning, the hundred small judgement calls that separate a senior build from a generic one. The client gets a senior all-day, every-day, instead of a senior for two hours and three juniors for the rest.

And yes, that's reflected in the invoice. AI-first delivery means the studio's leverage shows up as either a smaller bill, a shorter timeline, or a richer scope. Pick two, sometimes three. Never 'pay agency rates for AI-output quality'. That's not how this works.