How we use AI to build software, and where we do not trust it
AI makes our engineers much faster. It does not make them less responsible for what they ship. Here is how we use it day to day, and the decisions that always stay with a person.
Every engineer at Ugibo works with AI tools all day. It is a large part of why a first phase can arrive in weeks rather than months, and why a small team can build and look after systems that would once have needed a much bigger one.
It is also why we are strict about what those tools are allowed to decide. This note is how we work, so you know what you are buying.
Where it speeds us up
- Understanding your business We collect sample orders, spreadsheets, emails and process notes. AI reads them and drafts a first map of how work flows, which the engineer then corrects with the team.
- Writing routine code Forms, data models, connections to well-known systems and the plumbing every project needs. The engineer spends their time on the parts that are specific to you.
- Testing Generating test cases from real examples, including the odd ones people forget: the order with no delivery date, the customer with two accounts.
- Reviewing A second pass over every change, looking for mistakes, before a person reviews it too.
- Documentation Keeping handover notes and system docs current as the system changes, so nothing lives only in one person's head.
Where it does not decide
Some decisions are always made by an engineer who understands your business:
- The shape of the system. What talks to what, where data lives, what happens when something is down.
- Security and access. Who can see and change what. Nothing generated goes near credentials or permissions without a person reading every line.
- Anything touching money. Prices, invoices, refunds, tax. These are written and tested by hand against your real cases.
- What goes live. A person decides when something is ready and watches it the first time it runs on real work.
AI output that looks right is not the same as output that is right, and the difference is exactly what you hire us for.
The checks that stay human
Every change is read by an engineer before it is merged, whoever or whatever wrote it. Every system has tests that run on every change, and the tests that matter most (prices, permissions, the main path through each process) are written deliberately, not generated in bulk. And every system we look after has someone named as its owner, who can explain how any part of it works.
What you get from it
Faster first phases. More tests than a team our size would otherwise write. Documentation that is actually up to date. And more of the engineer's time spent with your team, understanding the work, rather than typing boilerplate.
What you do not get is software nobody understands. Every piece has an engineer who can explain it, and who will still be around to change it when your business changes.
Questions people ask
- Is code written with AI less reliable?
It can be, which is why every change is read by an engineer and covered by tests before it is merged, whoever wrote it.
- What does AI not decide?
The shape of your data, how the business rules work, and anything touching security or money.
Want to see how this works on a real process? Ask us to walk through a recent first phase, from the first visit to the day it went live.
Canonical URL: https://ugibo.com/blog/how-we-use-ai-to-build