The short version
Tools like Claude, Cursor and Copilot are extremely good at the boring half of programming: scaffolding, repetitive UI, boilerplate, basic tests, glue code. They are not good, on their own, at architecture, edge cases, security, or product judgement. So a sensible studio uses AI for the boring half and a human engineer for the half that matters. The 12-week build becomes a 4-week build, and the bill follows.
What the tools actually do
- Generate UI screens from a description. "Booking calendar with date picker, list of slots, deposit field." Done in minutes.
- Convert designs into working code. Figma to React Native, with live data wired in.
- Write integration code. Stripe, Firebase, Supabase, REST APIs, all stitched in by AI faster than by hand.
- Generate tests. The kind of tests humans write last and skip when busy.
- Explain unfamiliar code. Especially helpful when picking up someone else's project.
What the tools cannot do (yet)
- Decide what your app should be.
- Design a database that survives 3 years of feature creep.
- Catch security flaws that need understanding of your business rules.
- Hand a project to your team in a way they can run with.
These are the parts a human engineer keeps doing, on every project, every day.
Why agencies are slow to pass the savings on
If your billing model is "5 engineers at £750 a day for 12 weeks", AI tools that finish the same work in 4 weeks are an existential problem. Most large agencies have not adjusted yet. Some are quietly using the tools and pocketing the difference. A few small studios, ourselves included, have rebuilt the pricing around the new reality.
What this means for you
The same Android app that an agency would price at £40k will cost you £1,400 to £3,200 with us, with the same Kotlin underneath. The same iOS app that costs £80k at a London agency we ship for £2,800 to £4,500. See pricing.