# Machine speed, human signature: how we use AI without losing accountability

# Machine speed, human signature: how we use AI without losing accountability

Carlos Eduardo de Souza · 2026-10-14

AI drafts a lot in our pipeline. People still decide. Here is where AI sits in each stage, and where it does not.

In our pipeline, AI drafts a lot: flows, components, tests, translations. None of it ships without a person who signs it. This is how we split the work between machines and people, stage by stage.

### Speed alone is not the point

 AI makes the first draft cheap. That changes how a project feels: more options early, faster feedback, less time on repetitive work. It does not change who is responsible. Clients do not hire a model; they hire a team that answers for the result.

 So we set one rule before choosing any tool: every change that reaches a client has a named engineer behind it.

### Where AI sits in each stage

- **Discover.** AI summarises research and drafts questions. The team runs the workshop and writes the scope.
- **Design.** AI drafts many flows quickly. Designers keep the few that work and test them with users.
- **Build.** AI pairs on code and tests. An engineer reviews every diff before it merges.
- **Verify.** AI runs regression and accessibility scans. A person checks the delivery against what was agreed.

> The speed comes from the machine. The signature comes from a person.

### Measured, not assumed

 Every time we bring AI into a task, we record how long it took before and after, and what the review found. Some tasks get much faster. Some do not, and we stop using AI there. The numbers decide, not the hype.

```
task: component + tests
before: [hours]   after: [hours]
review: 2 changes requested
✓ signed off by a named engineer
```

### What this means for clients

 Projects start faster and reach working software sooner. Decisions stay with people you can talk to. And at the end of each cycle you see, in numbers, what AI changed in the work.

 Machine speed. Human signature. That is the line we work by.

## More from the journal

- [Journal](/blog.md)

- [Machine speed, human signature: how we use AI without losing accountability](/blog/machine-speed-human-signature.md): AI drafts a lot in our pipeline. People still decide. Here is where AI sits in each stage, and where it does not.
- [TYPO3, WordPress or headless? Choosing a CMS for a multilingual site](/blog/typo3-wordpress-or-headless.md): Three questions decide it: who edits, how many languages, and what your IT allows.
- [GEO: writing for people and for AI answers](/blog/geo-writing-for-people-and-for-ai-answers.md): Search is no longer only a list of links. What changes in content and code when answers come from AI.
- [React Native in 2026: when one code base is enough](/blog/react-native-in-2026-when-one-code-base-is-enough.md): Where cross-platform wins, and the cases where we still write native modules.
- [Verification and validation: how we check every delivery](/blog/verification-and-validation-how-we-check-every-delivery.md): Done is defined at the start. At the end, we check against it, item by item.
- [Working from Curitiba with teams in Vienna and New York](/blog/working-from-curitiba-with-teams-in-vienna-and-new-york.md): Hours that overlap, rituals that travel, and what we learned about working across time zones.
- [Our origin: from code partner for agencies to AI-first engineering](/blog/our-origin.md): Coopers started in 2013 as the team agencies called when they needed code. What we kept from those years, and what changed.
- [How we build a website, step by step](/blog/how-we-build-a-website.md): Five steps from the first question to the first weeks after launch, and what you see and approve at each one.
- [Monthly maintenance: why a website is never finished](/blog/monthly-maintenance.md): A site nobody looks after gets slower, less secure and harder to find. What monthly maintenance covers, and how we run it.

---
HTML: https://coopers.digital/blog/machine-speed-human-signature