All projects

Lama

2026

An AI agent that writes your tests for you

Founder & Engineer/Product
00:00 | 00:00

Writing automated tests is slow, repetitive, and breaks the moment your UI changes. The idea behind Lama was simple: what if you could just describe what to check, and an AI agent did the rest? That is what our team set out to build.

  • Tests from plain English. You describe a scenario, Lama gives you a real, runnable test.
  • An agent that uses your app like a person. It opens a real browser, finds the right elements, and works out the flow on its own.
  • Tests that heal themselves. Lama learns how your app is put together, so when the UI changes, tests adapt instead of breaking.
  • Code you own. It writes tests in the framework your team already uses, with nothing locked to us.
  • Where your team works. A desktop app and a browser extension, wired into the tracker you already use.
  • Built by a small team around one idea: testing should be something you describe, not script.

Where it started

Every team we had worked on hit the same wall. Tests took forever to write and broke constantly, so people spent more time maintaining them than shipping. We kept asking the same question: what if you could just say what you wanted tested?

How it works

You describe a scenario in plain language, like "log in and check the dashboard loads." Lama opens a real browser and clicks through it the way a person would, learning the pages and the flow as it goes. Then it writes clean, runnable test code in the framework you already use. Because it understands how your app fits together, it can repair tests when the UI shifts instead of simply failing.

The hard part

The interesting problems were never in the interface. They were in making an agent you could actually trust: getting it to understand a real app, remember it, and recover when things change. And making all of that feel effortless to someone who would rather not write tests at all.

The team

Lama was built by a small team that shared the same idea. We split the work across the agent, the product, and the marketing site, and shipped it as a research preview.