Insights

Healthy distrust: keeping people in charge of AI-written code

Developers themselves are more likely to distrust AI tools than to trust them. Here is how that caution becomes a working method.

The Stack Overflow Developer Survey 2025 reports that more developers actively distrust the accuracy of AI tools (46%) than trust it (33%).

It is the people who use these tools every day who say it. That is not a reason to stop using them. It is a reason to be clear about who is responsible for the result.

Distrust is a sensible default

An AI tool produces plausible work quickly. Plausible is not the same as correct. The sensible response is not to ban the tool or to believe it, but to arrange the work so that its output is always checked by someone who can be held to account.

How that looks in a project

  • A specification to check against. Requirements are numbered and each has acceptance criteria. A reviewer has something concrete to compare the code with, rather than a feeling that it looks right.
  • Review before acceptance. Changes are read by a person before they are accepted into the product.
  • Tests that can fail. Each requirement has a check. A tool cannot talk its way past a failing test.
  • Named accountability. After go-live there is an owner, a runbook and a support route.

What AI is good for here

Drafting documents and first versions of code, searching for faults, writing tests, and turning an agreed specification into working software faster. We use it for all of these. What it does not do is decide what is built, or sign anything off.

That is what we mean by AI-accelerated and human-governed. You can read the method in full on How we deliver.

Source

Tell us what you need built

Describe the idea, the problem, or the app you want rescued. We reply within 2 working days.