From 5% to 96% Test Coverage in a Day — Thanks to AI CoPilot

When I started this project and it is still a small codebase, I knew unit testing would be important, but like many developers, I pushed it aside after building the first suite of tests in favour of building out the core functionality. My main goal early on was to get the middleware logic working — the centrepiece of the app — and bring the Minimum Viable Product (MVP) to life. That meant skipping a good chunk of test coverage in the name of progress.

Once the MVP was working, I turned my attention back to testing. Looking at the project, I estimated at least a week or two of writing tests, depending on the complexity of each component. Some would be straightforward. Others — especially where edge cases or intricate branching logic were involved — would be far more time-consuming.

At that point, I turned to AI. I'd been using a “CoPilot” to help write code, and I figured it could also help with tests.

Early Missteps with AI

At first, I asked the AI to “write unit tests” without any real context or constraints. The result? A mess. The tests didn't work, they made incorrect assumptions about the code, and the whole effort turned into a loop of broken test code and frustrating rewrites. The AI generated a lot of code — but not the kind I could use without serious refactoring.

Giving AI Direction

The breakthrough came when I shifted my approach. Instead of treating the AI like a magician, I started using it more like a junior developer. I gave it one file at a time, explained the intent of the code, and told it to focus on writing accurate, functional tests for that file only.

That changed everything.

Suddenly, the tests were easier for the AI to implement. They were scoped correctly, aligned with the actual logic, and didn't require hours of debugging. My AI worked it's way through file by file, writing and running tests.

The Results

In less than a day, test coverage jumped from 5% to over 96%.

• ✅ Statements: 96.72%
• ✅ Branches: 95.67%
• ✅ Functions: 92.1%
• ✅ Lines: 96.87%

Tests that would have taken days to write were completed in hours, and the peace of mind that comes with nearly full test coverage? Invaluable.

I went from one Test Suite and 5 tests to;

• 20 Test Suites
• 136 Tests passed

What I Learned

• AI isn't magic — but it's an incredible assistant. Give it context, guide it, and you'll get great results.
• Focus matters. Asking the AI to “test everything” leads to chaos. Asking it to test this file, this function, this case— that works.
• Expect to debug — but way less than writing from scratch. The AI sometimes makes mistakes, but the overall process is much faster than going solo.

Whether you're kicking off a new project or trying to backfill tests on an older one, AI can be a massive time saver — if you use it well.

Are you using AI for writing tests in your projects? What's been your experience?

For us, this same test-first discipline is essential in production products like PrimeCRM, where reliability and governance have to scale with the business.