All writing

Rebuilding an old system, with AI

Guide3 min read

Rebuilding an old system, with AI

Guide · November 21, 2026 · 3 min read

Most established businesses have one: a system built ten or twenty years ago that still runs something important. The people who wrote it have moved on, the documentation is thin, and every change is a risk. It works, so nobody touches it, and the business slowly bends itself around what it cannot do.

Rebuilding such a system used to mean a long, expensive project with a frightening switch-over at the end. AI has not removed the risk, but it has changed where the time goes, and made a careful rebuild realistic for businesses that could not have afforded one before.

Why old systems are hard to replace

The difficulty is rarely the technology. It is that the old system is the only complete record of how the business actually works. Years of exceptions, special cases and quiet fixes live in its code and nowhere else. Rebuild it from what people remember, and those rules are lost.

What AI changes

  1. Reading the old code. AI can read large amounts of old code, in languages few people still write, and explain what each part does in plain language.
  2. Recovering the rules. It can pull out the business rules buried in the code, such as how a price is worked out or when an order is held, and set them down for the people who run the business to confirm.
  3. Documenting the data. It can map an old database, table by table, and describe what each field really holds, including the ones whose names no longer mean anything.
  4. Writing the tests. It can turn the old system's behaviour into tests, so the new system can be checked against the old one, case by case.
  5. Moving the data. It can help write and check the scripts that carry years of records across, and find the records that do not fit.

What AI does not do is decide. Every rule it finds is confirmed by someone who knows the business, and every line of the new system is reviewed by an engineer.

Rebuild in pieces, not in one go

The safest rebuilds never have a switch-over day. The new system grows beside the old one, taking over one part at a time: first the reports, then quoting, then scheduling, each moved only when it has proved it does the job.

For a while both run side by side on the same work, and their results are compared. When they agree for long enough, that part of the old system is retired. If they disagree, the old system is still there, and the difference is a rule nobody knew about, found before it could do harm.

Better than the original, not a copy

A rebuild is the moment to drop what nobody uses and fix what everyone works around. The test is simple: every rule in the old system is either kept on purpose or dropped on purpose, never lost by accident.

It is also the moment to build in what the old system could not have: data in one place that AI can work from, agents that take on the routine work, and connections to the tools the business uses today.

How a rebuild runs

StageWhat happensWhat you get
ReadAI and engineers read the old code and dataA plain account of what the system really does
ConfirmThe business checks every rule that was foundA specification you sign
RebuildThe new system is built in pieces, tested against the oldEach part proven before it takes over
Run togetherOld and new run side by side on real workDifferences found while they are still harmless
RetireEach old part is switched off once the new one has proved itselfOne system, owned and documented

Questions to ask before you start

  1. What would it cost the business if the old system stopped working tomorrow?
  2. Who still understands it, and for how much longer?
  3. Which parts does the business actually use, and which could go?
  4. Can the old and new systems run side by side for a while?
  5. Who will confirm the rules that are found?

An old system that still works is an asset: it is the most complete description of your business that exists. Rebuilt carefully, it becomes the foundation for everything that comes next.

Start with one problem, or bring us the whole operation.