AI agents migrate the code, and parity tests prove it matches the old system, which keeps running.
Legacy keeps serving the rest
Each subsystem is its own goal with acceptance criteria. The legacy system keeps serving everything that hasn't moved, so there is never one weekend where everything switches at once.
Old and new outputs compared
Parity tests run the same inputs through the legacy code and the new code and compare the output. Traffic moves only when they match and the review agrees.
A team, or a price per piece
A forward deployed team fits a whole migration, and a subsystem with a clear edge can also be a fixed-price project on its own. The free audit maps the subsystems and the order to move them.
The legacy system serves everything that hasn't moved, and the next subsystem runs the same loop.
A big-bang rewrite bets the business on one switch date. We move one bounded subsystem at a time behind a strangler pattern, so the old system keeps serving everything that hasn't moved and every switch is small enough to reason about.
See the evidence →Your team keeps shipping features while we migrate. Each subsystem ships on its own and the first pieces ship in weeks, so progress shows up as subsystems moved, not as a percentage on a status deck.
See how we price →A discovery session maps the repository and its conventions into Steer's Context, and when a rule lives only in someone's head, Steer interviews them and writes it down. Parity tests then pin the old behavior, so the knowledge outlives the migration.
See spec-driven development →✳ The audit is free. It maps the subsystems, the seams between them and the order to move them, and that map sets the size of the team or the price of a first subsystem.
A discovery session reads the repository, its conventions and its seams, and stores the map in the project's Context. Interviews with the people who know the edge cases fill in what the code doesn't say.
Every subsystem is split into goals with acceptance criteria, and one criterion is always parity: the same inputs through the legacy code and the new code give the same output.
Claude Code agents do the migration on branches in your GitHub, each run in a fresh isolated sandbox that is thrown away afterwards, while our forward deployed engineers steer them.
An independent review model and a forward deployed engineer judge each pull request against the spec. The evidence report shows the tests that ran, including the parity run, case by case.
Traffic moves to the new code one subsystem at a time behind a strangler pattern. Changes that touch production data ask a person before they merge.
The same loop runs for every subsystem, so the fifth piece is as well evidenced as the first, and your team can see each one in the workspace.
Every legacy modernization engagement includes:
2 weeks
Subsystem by subsystem instead of a big-bang rewrite. Zero post-handoff incidents.
3 modules
We run Steer on ourselves first: invoicing, projects and allocation moved off our old system one bounded piece at a time, each specced, evidenced and approved before the next.
Pento forms a pod of forward deployed engineers for your work, for example four engineers or whatever machine learning, software, DevOps and MLOps talent you need. They work with Steer alongside, and the fee depends on the pod's size.
For integrations and smaller projects you can scope up front. After the free audit we write the scope with you as goals with acceptance criteria and fix the price. Evidence on every pull request shows when each goal is done.
Not covered here? Bring the question to a discovery call. An engineer reads every request.
Book a discovery callContact us
Tell us what the legacy system runs and what is stuck behind it. The audit is free: we map the subsystems and the order to move them, then quote a team for the migration or a fixed price for the first subsystem. An engineer reads every request.