PEQY mascot

Copilot reviews AL? We measured it.

GitHub Copilot is a fine reviewer for the languages it saw a lot of. AL isn't one of them. And on an ERP, a plausible-sounding wrong finding costs real review time.

What we measured

We ran GitHub Copilot's code review on real Business Central pull requests at a BC partner and checked its output:

  • Every compile error it claimed was refuted by the AL compiler. Every one we checked, without exception.
  • Roughly half of the findings the developers engaged with didn't hold up on inspection.

One of the partners who came to us has since switched Copilot review off entirely.

To be fair: on mainstream languages, with plenty of training data, Copilot review behaves much better. AL is the edge case, unless AL is your whole business.

Real bugs, real reviews

We forked bc2adls, a community BC tool many partners know, and replayed two feature commits exactly as they were merged upstream. Each contained a bug that was only fixed months later. PEQY reviewed the PRs cold.

Disabled tables kept exporting

In a roughly 1,000-line feature diff, PEQY flagged that StartExport() never checks the per-table Enabled flag, so tables a customer disables still get exported. The same bug shipped upstream in 2022.

Read the review and the upstream fix.

An errored run silently skipped records

PEQY flagged that checkpointing the highest flushed timestamp on an errored run makes the next run skip records that were never exported. Upstream fixed the same bug in 2026.

Read the review and the upstream fix.

“Do you basically let a butcher test his own meat? If I let Claude write the code and let Claude review the code, would I ever find any issues Claude is going to miss?”
René Brummel, freelance BC developer, on reviewing AI-written AL with the model that wrote it

What an AL-native reviewer does differently

  • It proves its claims before making them. Every compile error PEQY claims is verified against the actual AL compiler first.
  • It knows what the symbols mean. Findings resolve against the Business Central base app, system app and platform layers instead of a guess from the diff.
  • It runs where BC partners actually work. GitHub, GitLab and Azure DevOps, including on-premises Azure DevOps Server and TFVC, which bundled reviewers don't reach.

See it on your own AL

Point PEQY at a repository and let it review your past pull requests. You judge the findings.

Start free trial