AI writes the code. Who owns the work?
July 2026 was a busy month for developer tooling. But for companies the real issue is not which model writes better code; it is who carries responsibility for the code that gets produced.
A short recap
In July every major model provider shipped a new release. GitHub Copilot added its first open-weight coding model. On the enterprise side, AI products that live inside the workspace were announced one after another. What the tools can do is no longer the debate.
The debate is this: if a team accepts thousands of generated lines a week, who is reading them?
Speed went up, review capacity did not
The bottleneck in software has not been writing code for a long time. The bottleneck was deciding what to build, verifying what was built, and shipping it. AI tools accelerated the easiest part and left the hard parts where they were.
In practice: a developer's day of output used to take a colleague half an hour to review. Now the same developer produces three times as much, and review time has not moved. The difference goes somewhere, usually into testing and production.
AI tools increase a team's output only if review and testing capacity grow alongside. Otherwise speed turns into debt.
Three practical consequences
First, when choosing a supplier, "do you use AI" has lost its meaning; everyone does. The useful question is: how do you verify what gets produced? What is your test coverage? When a bug reaches production, how many minutes until you can roll back?
Second, data boundaries need to be explicit. If your codebase is sent to a model, which model, which provider, under what retention policy? That belongs in the contract now.
Third, ownership. If no human owns the maintenance of AI-generated code, that code becomes the region nobody wants to touch six months later. Tools change; this does not. Every module needs an owner.
What we do
We use these tools heavily on our own projects. But three things stay fixed in everything we deliver: automated tests on critical paths, releases that can be rolled back, and a named owner for every module. Tools are accelerators; responsibility cannot be delegated.
Frequently asked
- Is AI-generated code safe?
- What matters is the verification process, not the origin of the code. In a process with automated tests, review and a rollback mechanism, AI-generated code is safe; without those three, human-written code is not safe either.
- What should I ask a software supplier?
- Ask how they verify generated code, what their test coverage is, how quickly they can roll back a release, and which AI providers your codebase is sent to. Those four questions reveal a supplier's maturity directly.
- Do AI tools reduce the need for developers?
- They shorten writing time but do not reduce the burden of deciding, verifying and maintaining. Teams usually do not shrink; the centre of gravity moves from producing to reviewing.