In 2019 I started a studio with one colleague. A year later we were more than ten people, and serving the clients we had took every one of them.

In 2026 one engineer with technical depth, good communication and AI tools can deliver work that used to need more people and more time. Not all of it. But a lot more than most founders hiring an app team have priced in.

Here's what changed, what didn't, and what I'd look for if I were hiring a mobile team today.

0+

people it took to serve serious clients in 2019

Affarit Studio, one year after two of us started it

What a harness is

A model on its own is a very clever autocomplete. A harness is everything around it that makes it useful on a real codebase: the instructions it reads every session, the files it may open, the tools it can run, and the checks its work has to pass before anyone trusts it.

On a mobile project that means the architecture rules, a map of the modules, how to run the tests on both platforms, and a short list of things nothing touches without a human: signing configs, the release pipeline, the payment flow.

A senior engineer builds the harness once and keeps sharpening it. Then a lot of the typing moves to the model, and the engineer's day moves to deciding, checking and owning the result. Since March even the content side of my own business runs this way: a repo, a rules file the agent reads every session, and Claude Code doing the drafting.

One feature, one engineer
  1. 1Spec the problem
  2. 2Agent drafts small diffs
  3. 3Tests and review
  4. 4Engineer ships it

What got cheaper

Basic coding. For everyone. A screen built from a design, a repository wired to an API, the scaffolding around a new module, a first round of unit tests: a good harness drafts these fast, and they are cheap now.

The work around the code got cheaper too. One person can now prototype, document, record a demo and automate parts of delivery. The borders between those jobs are blurring.

That has an uncomfortable side for anyone selling hours of basic coding: when everyone can produce it, it gets more competitive and cheaper. If that's what you're buying, you're overpaying; if that's what you're selling, the model is your competitor.

What didn't

Before

Cheaper every month

  • The first version of a screen
  • Boilerplate, wiring, scaffolding
  • A first pass at tests and docs
  • A demo by Friday

After

Still expensive

  • Expertise: knowing what breaks on real devices, with real users
  • Judgment: seeing that a simpler, cheaper solution solves the problem
  • Responsibility: someone who says it first when something slips
  • A working result: in the store and used, not just compiled

A client doesn't buy code; they buy a result. AI made the code part cheaper, so the result part is now most of what you're paying for. That's also why hourly rates mislead: compare what each option ships, and who answers the phone when it breaks.

Where the harness earns its keep on mobile

  • Turning a design into a Compose or SwiftUI screen, with every state the design forgot to draw
  • Carrying an API change through the data layer, the mappers and the tests
  • Writing the regression test for the edge case you found this morning
  • Keeping strings, analytics events and release notes in step with the code

And where it doesn't: the crash that only happens on one manufacturer's phones, the store review that rejects a flow for a reason nobody wrote down, the payment edge case a client mentions on a call. Those still need someone who has seen them before.

What this means if you're hiring an app team

  1. 1Hire for judgment firstAsk about a time they talked a client out of a feature. The answer tells you more than a coding test a model could pass in a minute. You're hiring the part of the job the model can't do.
  2. 2Ask to see the harnessHow do they give an agent context? Where do specs and plans live? What has to pass before anything merges? A senior engineer has concrete answers to all three, and can show you the files.
  3. 3Make one person own the resultNot the ticket, the result. Someone who tells you on Tuesday that Friday is at risk, instead of telling you on Friday.
  4. 4Start smaller than you thinkOne senior engineer with a good harness, a designer, and a founder who decides. Add people when the work proves you need them, not when the funding says you can.

Notice that none of these is about which AI tool they use. Tools change every few months. Judgment, and a harness built around it, compounds.

AI is leverage. It multiplies what an engineer already knows. It doesn't replace knowing it.

What I tell founders who ask how many developers they need

The question for 2026 isn't how many developers you can afford. It's whose judgment you're paying for, and whether their harness lets that judgment reach every line that ships.

Building with AI?Let's talk about your appBook a free intro call