An Old System Nobody Dares to Touch
The system works. Its author left long ago, the current vendor barely knows your business, and the documentation is missing or out of date. Any change might break something, and nobody knows what.
What Do We Do About It?
It happens in a small company with a 15-year-old system and in a bank with a 30-year-old platform. Sorting it out used to be so expensive that only large organizations could afford it. Now it costs noticeably less.
First: the checks
Before changing anything, we write down how the system should work as automated tests. We write them in business language, not developer language, so your specialist can read them and say: "Yes, that's how we calculate it."
Then: the changes
Then we develop the system in small steps. After each step the tests run again, and if something breaks, you see it immediately, not a month later from a user. There is no need to rewrite everything from scratch.
So Where Does AI Come In?
The slowest and most expensive part of this work used to be understanding someone else's code. A developer could spend weeks reading it, trying to figure out what it does and why.
Since 2025, AI agents read and explain large codebases much faster. What took weeks now takes days. So even a small company can afford to sort out an old system. But how the system should work is still decided by your specialist.
Where Does This Already Work?
We have looked after some systems for more than thirty years. Among them are a utility billing system in Omsk with 500,000+ accounts and a bank CRM with 7,000 users. The approach is the same for 50 users or 50,000.
