How we build with AI – and where we keep it out
· 3 min read · Flashport
AI has made writing code cheap. It has not made it cheap to make a shop fast, an upgrade safe or a payment correct. This is how we split the work.
Where AI makes us faster
- First drafts of code and tests
- Boilerplate, migrations and documentation
- Prototypes that let us test ideas early
- Code review as an extra pair of eyes – never the only one
What a person always writes and checks
Anything that moves money, touches personal data or decides performance is written and checked by a senior engineer who can explain every line.
An example from one of our platforms: we never retry a bank payout automatically. The bank's API does not detect duplicate orders, and a timeout does not tell you whether the money has left. A well-meant retry can pay someone twice. So there is exactly one attempt, the status "unclear", and a check against the bank before a person decides.
Our rules
- A senior engineer reviews every change.
- We measure before and after instead of guessing.
- Tests first where money, data or key flows are involved.
- No client data in AI tools without consent.
What this means for you
You get the speed AI brings and the accountability only people can take. We measure ourselves by results: a faster shop, a safe upgrade, a payment that is correct.