# Check AI-written code before your customers do

> Teams ship more code than anyone reviews line by line. Checking each deploy from the outside, as a customer would, catches what review and unit tests miss.

By Jason Nguyen, Founder, Testsy. Published 2026-11-21. https://testsy.dev/blog/check-ai-written-code-before-customers-do/

## More code, same number of reviewers

Coding assistants have changed how much code a team ships. Google's 2025 DORA report, from a survey of nearly 5,000 technology professionals, found that AI adoption now has a positive relationship with delivery throughput, and that it [continues to have a negative relationship with software delivery stability](https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report). More gets shipped; more of it breaks.

Review does not scale with it. A reviewer reading a 600-line diff written in two minutes checks that it looks reasonable, not that the checkout still works. Unit tests written in the same session as the code check the code's own assumptions.

## Check the behaviour, not the diff

The part of quality that does scale is checking what the customer gets, after every deploy, without reading the diff:

1. Keep a short suite of the journeys that earn money or would embarrass you: sign-up, sign-in, the core create-and-save, checkout.
2. Run it on every deploy, against the deployed address.
3. Fail the pipeline on a broken journey, with the replay attached.

This does not depend on who wrote the change, a person or an agent. It also does not depend on the change being small enough to review well.

## What that looks like with Testsy

The suite is plain text, so the person who knows the product can write and review it:

```text
Sign in with the stored account "member".
Create a task "Release smoke sample" in the project "QA".
Reload and check the task is still listed with that title.
```

It runs from [GitHub Actions](https://testsy.dev/blog/run-browser-tests-after-every-deploy-with-github-actions/), [GitLab](https://testsy.dev/blog/run-plain-language-tests-in-gitlab-ci/) or [any CI](https://testsy.dev/blog/run-browser-tests-from-any-ci-with-one-file/), and every result keeps each check's value, the screen it was read from, the replay and the video.

![The Runs page: every run with its verdict, filtered by verdict, type and where it was started.](https://testsy.dev/docs/img/runs-light.webp)

## Give the agents the same tests

If your developers use coding agents, let the agents run the suite too, over [MCP](https://testsy.dev/docs/mcp/), before they open a pull request. The agent reads the failed check and fixes its own change; the pull request arrives with a passing run linked. The tests do not change between the agent's run and the pipeline's, so the agent cannot pass by testing something easier.

## What it will not catch

A browser suite does not review code quality, security or performance. It catches the broken journey, which is the failure customers notice first. Pair it with code review focused on design and risk, and with your monitoring.

This post is part of [Release testing for product teams](https://testsy.dev/blog/release-testing-for-product-teams/).
