If AI writes the code, what are you paying a studio for?

Every founder should ask a studio this question, point blank: “if AI writes the code, what am I paying you for?” It's fair. If code got cheap, it makes sense to be skeptical of anyone still charging to build software.

The best answer I've read so far isn't mine. It's from Gergely Orosz, author of The Pragmatic Engineer, in a piece published on January 6, 2026 and opened to non-subscribers in September. He clearly separates the engineering skills that lose value from the ones that gain value when AI writes almost all the code.

What AI already does well, per The Pragmatic Engineer

Orosz estimates that about 90% of the code he writes today, in his stack, comes from AI. From there, he lists what loses value:

• Prototyping. Non-technical people already build prototypes in tools like Lovable and Replit, but the prototype still lacks the 30% missing from an MVP.
• Specializing in one language. AI helps anyone work in a codebase they don't know.
• Splitting front end and back end. One person with AI covers the whole stack.
• Implementing well-defined tickets. Given a clear task, AI delivers something usable.
• Manual refactoring and typing code. Describing the change is now faster than doing it by hand.

One data point helps size this. Orosz cites a 2025 Atlassian survey of 3,500 respondents: developers spend 16% of their week writing code. Even before AI, typing was never most of the job.

What gains value when code gets cheap

The other list is the one that answers the founder's question. According to Orosz, these gain value:

1. Writing precise specs. Tickets that cover performance, reliability and what happens when something goes wrong.
2. Tests and test infrastructure. Coding agents need feedback: compilers, static analysis, automated tests.
3. Product sense. Deciding what to build matters more than building fast.
4. Architecture decisions. If the structure isn't defined, AI can produce a system that's hard to maintain.
5. Managing tech debt. More generated code means more debt to track.
6. Reliability, performance, scale and security. They need someone who knows what to verify and when to write by hand.
7. Verification. Being a real engineer, not just someone who produces code.

My favorite line of his: “software engineering fundamentals should become more important, the more a team relies on AI to generate code.” The more code AI generates, the more problems there are to catch.

And one number shows the other side. The Cortex 2026 benchmark, covering more than 50 engineering leaders and cited in the piece, found a 30% increase in change failure rate. More code, faster, breaks more.

What you buy when you hire a studio today

Translating the list into what a founder needs: when you hire a studio in the AI era, you're not paying for lines of code. You're paying for:

• A spec that fits your budget. Deciding what goes into the first version and what waits. This is where most projects blow up.
• An architecture that survives version two. The product will change. The structure has to allow it.
• Security verified, not assumed. Access rules, keys in the right place, protected data. It's worth knowing whether AI-written code is secure.
• Someone accountable for the result. Price, deadline and quality with a name and a contract.
• Design that drives usage. Pretty screens aren't enough; the flow has to get users to value.

Karri Saarinen, cofounder of Linear, puts it this way in Orosz's piece: “directing and managing agent work becomes the craft.”

The risk of paying only for code

The old hiring model charged by the development hour. With AI, that model gets awkward for both sides. The vendor delivers faster and doesn't want to charge less. The client pays for hours without knowing whether they were spent thinking or typing.

Worse: a vendor that only generates code, without handling architecture and verification, delivers exactly what Orosz warns about. A system that works today and is hard to maintain tomorrow. At the Studio, architecture and verification are part of the service. That's how AI Vou Eu, an AI travel SaaS, went live in 4 weeks.

How we do it at the Studio

We use Claude Code and Figma for design and build. AI writes a lot of the code. The team and I decide what it writes, review what goes in and own the result.

In practice, that becomes three commitments:

• Fixed price after discovery. You pay for a live product, not by the hour.
• Deadline written into the contract. Like LCX, which merged one system per country into a single system, live in 29 days.
• Support after delivery. Every system ships with an AI agent that runs daily, flags failures, suggests improvements and checks that the data makes sense.

With AI writing fast, project time shifts: more time on scope, architecture and review, less time typing code.

AI changed the price of writing code. It didn't change the price of getting it wrong. If you need a custom system or SaaS live, with someone answering for the decisions AI doesn't make, that's what the Studio delivers.