We are building the review layer for AL & Business Central.

I am a software engineer. I have built things at startups and at Google, and when LLMs arrived I wanted to mix the two worlds I knew: the deterministic one, where scripts do exactly what you tell them, and the new one, where models read and write code. The first idea was fixing technical debt with LLMs: test coverage, refactoring, modernization. It half worked. Getting code changes was easy. Getting the right code changes was a very different beast. That is how we ended up in code review.

Then Münster did the rest. A couple of companies here build on Microsoft Dynamics 365 Business Central, and they were not happy with Copilot reviewing their AL: it kept making things up, claiming code will not compile when it compiles just fine. They asked: you do code reviews, can you do that for AL? We said: let us find out. Apparently we can. They have since switched Copilot review off in their repositories and rely on PEQY.