AI is a co-pilot, not a replacement
The honest version of this post: AI has changed how our engineers spend an hour, and it has not changed what we are accountable for. A generated function nobody understands is a liability with good formatting.
Where it earns its place
- Scaffolding. Boilerplate, migrations, test fixtures, the tenth variant of a form.
- Translation. Moving a legacy stored procedure into readable application code, with a human checking the edge cases.
- Review. A first pass over a diff that catches the obvious before a colleague spends attention on it.
- Documentation. Turning a merged pull request into a paragraph someone can actually read.
Where we do not use it
We do not let generated code near authentication, payment handling or access policies without a line-by-line read. Not because the output is usually wrong, but because "usually" is the wrong standard for the parts of a system that fail quietly.
What actually changed
Throughput went up somewhere between fifteen and thirty percent depending on the project. More interestingly, the shape of the work changed: less typing, more reviewing, and a much lower tolerance for code nobody can explain.
The teams that got the most out of it were the ones with the strictest review culture before they started. That is not a coincidence — AI amplifies the process you already have, in whichever direction it already points.
Part of the team building and running the products behind these posts at Hedaya Global Solutions.
