AI writes a lot of our code. Here’s exactly what that means for you.

We call ourselves an AI-powered software studio, which by now is close to meaningless — everybody says it, and it usually means somebody on the team has a Copilot licence. So here is the specific version, including the parts that don't flatter us.

The industry context first, because it's genuinely striking. Developers report that 42% of their code is AI-generated or assisted. Google said in April that 75% of new code at the company was AI-generated and approved by an engineer. Gartner expects 60% of all new code to be AI-generated by the end of this year. A quarter of the YC Winter 2025 batch had codebases that were 95% AI-generated.

So yes. AI writes a lot of our code too. The interesting question isn't whether — it's what that should mean for the person paying.

What it changes for you

Speed, mostly at the start. The scaffolding of a project — data models, forms, CRUD screens, test fixtures, the third variation of a layout — is where generation genuinely shines. Work that used to take a fortnight takes days. That's real, and it's reflected in what we charge rather than pocketed.

More iterations before you commit. When a working slice is cheap to produce, you get to see three versions of a screen instead of approving a static mockup and hoping. This is the underrated benefit. Most software disappoints not because it was built badly but because the wrong thing was specified confidently.

Nothing about the hard parts. Architecture, data modelling, security, what happens when two people edit the same record — the decisions that determine whether software survives its second year. Generation is not close to autonomous there, and anyone telling you otherwise is describing a demo, not a maintained system.

The bottleneck was never typing. It was knowing what to type, and noticing when what came out is subtly wrong.

The part nobody advertises

AI-generated code has a characteristic failure mode: it is plausible. It compiles, it passes the happy path, it reads like something a competent developer wrote. And then it does the wrong thing on the edge case, or quietly reintroduces a bug you fixed last month, or handles an error by swallowing it.

That's a genuinely harder review problem than obviously-bad code, and it's why the same surveys reporting high adoption also report rising anxiety about code quality. Volume went up. Scrutiny didn't automatically follow.

So our rules are boring, and deliberately so:

• Everything is read by a human before it merges. All of it, line by line.
• Generated code is held to the same bar as written code — if we can't explain it, it doesn't ship.
• The hard parts are designed first, by a person. Schema, auth boundaries, failure behaviour. Then we build.
• You get the repository either way. Ownership doesn't change because a machine helped.

Why we tell you this

Two reasons, both self-interested enough that you should believe them.

The first: it's checkable. You get the source code and the commit history. If we claimed everything was lovingly hand-typed, the repository would say otherwise within about a minute.

The second: the alternative positioning is worse. Studios currently split into those pretending AI isn't involved and those implying it does everything. The first is charging you for typing. The second is going to hand somebody a codebase nobody understands, and that bill arrives in year two.

We think the honest middle is where the actual value is: AI for the volume, senior judgement on the parts where being wrong is expensive, and complete transparency about which is which. It's the same reasoning behind Lilium Engine, where we build agents for clients — we'll tell you the tasks worth handing to a model, and the ones where you want a person and a checklist.

If you'd like a specific answer about your own project — where generation would genuinely save you money and where it wouldn't — that's what a first call is for. It's free, and we'll happily tell you when the answer is "this doesn't need custom software."

---

Sources: State of Code Developer Survey 2026, Sonar · How much software is written by AI in 2026 · AI code generation statistics 2026