I didn’t start building vincent because I wanted to create another AI coding tool.

I built it because I got tired of being the workflow engine myself.

For many coding tasks, especially QA issues, I kept giving agents essentially the same process:

1. Understand the ticket.
2. Check whether the issue could be valid.
3. Reproduce it.
4. Fix the root cause.
5. Add regression tests.
6. Update the documentation.
7. Bring everything back for review.
8. After approval, open the PR and update the ticket.

The problem wasn’t that agents couldn’t perform these steps.

The problem was that I had to keep orchestrating them.

I was carrying context between stages, checking whether conditions had been met, deciding when the workflow should stop, and repeatedly explaining what should happen next.

At some point the obvious question became:

Why am I describing this process again?

Why can’t I describe it once and simply provide the ticket ID next time?

That question became Vincent.

The goal is not to hide everything inside one huge prompt. It’s to make agentic development workflows explicit, reusable and inspectable with control flow, checks, persisted outputs and human gates when human judgment is actually valuable.

I wrote the story behind it here:

https://lezli01.is-a.dev/vincent/docs/why/the-workflow-i-kept-repeating-and-how-vincent-was-born.html

I’d be interested to hear how others are handling this: are you still orchestrating coding agents manually, or have you started turning your recurring processes into reusable workflows?