How we work: craftspeople in the age of AI
Zebup is a Montreal software development studio that uses AI every day to write code, and answers for every line it ships: human review, automated tests and decisions written down.
The craft
We use AI every day to write code. What we never hand over to it is the decision about what to build, and the verification of what was built.
The tool changed.
The craft didn't.
A craftsperson isn't someone who refuses machines. It's someone who answers for their work: they know their materials, they decide what needs doing, and they check before they deliver.
AI writes fast. That makes review and testing more important, not less. At Zebup, AI is a tool in the hands of someone who can recognize good code and owns the result.
Five practices, one concrete example each
These are everyday habits, not posted principles. The examples below illustrate how each one plays out on a project.
-
Automated tests on every delivery
No feature without a test that protects it. That is what lets you change the code six months from now without fear.
An example Before changing a price calculation, a test describes the expected result for every known case. If a change breaks one, we see it before you do.
-
Decisions written down before the code
A short design document for every subject that matters: what we do, what we don't, and why. It stays in your repository.
An example Before building the sync with your accounting, a short document says what is synced, in which direction, and what happens on error. It stays in your repository.
-
Review as the safeguard
Every line of code is read by a human before it ships, including what AI generated. Especially that.
An example A function generated by AI goes through the same review as one written by hand: someone reads it, questions it and refuses to ship it until it is understood.
-
Code that belongs to you and can be picked up
Readable, documented, with no dependency on us. Your team, or another provider, can carry on without calling us.
An example A new developer, from your side or elsewhere, can set up the project by following the repository guide and run the tests on day one.
-
Ship small, ship often
Batches of a few weeks, each one in production and verifiable, rather than one big reveal at the end.
An example Instead of a full portal delivered at the end of the project, a first batch puts login and the case list online. You see it working before the rest exists.
AI, plainly
We use AI tools openly, every day. Here is what that means, and what it does not.
What it does for us
- Write a first version of the code, which we read and correct.
- Suggest improvements to existing code.
- Draft tests and documentation.
- Quickly explore code we do not know yet.
What we never hand over to it
- The decision about what to build.
- The judgment of what is correct.
- Responsibility for what ships.
How we verify
- All code is read by a human before it ships, including what AI generated.
- Every delivery is protected by automated tests.
- Important decisions are written down before the code.
AI writes fast. That is what makes review and testing more important, not less.
Frequently asked questions
Is your code written by AI?
In part: we use AI every day to write first versions. But every line is read by a human before it ships, and every delivery is protected by automated tests.
Who checks the code AI generates?
A human, always. The review is the same as for hand-written code, and automated tests confirm that the expected behavior is there.
Why write decisions down before the code?
Because a short document that says what we are doing, what we are not doing and why prevents misunderstandings, and it stays in your repository for whoever picks the project up.
What do we receive at the end of a batch?
A version in production that you can verify, readable and documented code, tests that protect it, and the decisions made, written in your repository.
Can we take the project over without you?
Yes, that is the goal. The code is readable and documented, with no dependency on us: your team, or another provider, can carry on.
Let's talk about your project
The first conversation is free, with no commitment. We reply in English or French.