Deterministic recipe engines that AI agents can drive are the best choice for code migration across many repositories, because they make the same change the same way every time. For a single app in an IDE, an LLM agent is quicker to start, and for mainframe or cloud moves a dedicated modernisation service fits better.
The bottom line on code migration agents
An agent that writes code and an engine that rewrites code are not the same thing, and the difference decides whether your migration lands in a week or drags for a quarter. A language model is brilliant at the messy last mile: the odd custom wrapper, the test that needs rethinking, the comment that's now wrong. It's poor at doing exactly the same edit four hundred times without drifting.
So here's where I'd land. If you're moving a whole estate through one upgrade, such as a Java version, a framework major or a logging library swap, pick a rule-based engine with ready-made recipes and let an agent orchestrate it. If you're upgrading one application and a developer will sit with it, an IDE or CLI agent is fine and far less setup. If the job is mainframe, Windows or VMware estates heading to the cloud, look at the hosted modernisation services built for that, because they bring assessment and verification a general coding agent won't.
And if you need to ship the change as pull requests across hundreds of repos, you need campaign tooling on top, whatever writes the code.
Ten migration tools, from rule engines to agents
AWS Transform
Amazon's hosted agentic service for assessing and modernising whole estates, with mainframe, Windows and VMware workloads all in scope.
GitHub Copilot modernization
For Java, .NET or C++ stacks heading to Azure, this agent plans upgrades from a CLI and carries them out in the IDE.
Sourcegraph Batch Changes
Opens and tracks pull requests across many repositories and code hosts at once, giving large engineering orgs a campaign view of sweeping changes.
Moderne
A commercial platform that runs deterministic OpenRewrite recipes across an entire code estate and lets AI agents call them.
OpenRewrite
Open source refactoring engine with ready-made recipes for framework migrations and security fixes, run inside your own build.
Codemod
Compiler-aware codemods meet campaign tracking here, with an open source engine for single repos and a hosted platform for org-wide migrations.
GritQL (Grit)
Declarative query language and CLI for searching and rewriting source code, written with large-scale migrations in mind.
Mechanical Orchard Imogen
Mainframe teams use it to rewrite legacy systems into modern, cloud-ready code, with automated verification built into the process.
ast-grep
A fast, polyglot command-line tool that matches syntax patterns rather than plain text, for structural search, linting and rewrites at scale.
Konveyor
Open source project for analysing legacy applications and re-platforming them onto Kubernetes and cloud-native setups, now with an AI component.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Why rules still do the heavy lifting
A recipe is a small program that understands your code's syntax tree. It finds every call to a deprecated method, rewrites it, adjusts the imports and leaves formatting alone. Run it twice and you get the same diff twice. That's what lets a reviewer approve a change across hundreds of repositories without reading every line: they review the recipe once, then spot-check the output.
An LLM agent, by contrast, reasons about each file fresh. That's a strength on code nobody wrote a rule for, and a weakness at scale, because small variations creep in and every pull request becomes a new review. The setup worth looking for combines the two: the agent decides which recipe to run, runs it, then handles whatever the recipe couldn't. If you're picking a general coding assistant for that last-mile work, this list of AI coding agents covers the field.
One housekeeping point if you rely on the open source recipe ecosystem: OpenRewrite's docs say recipes are moving from Maven Central to the Code Genome Project, so check your build can still resolve them.
Estate, app or mainframe: three different jobs
Estate-wide upgrades are a coordination problem as much as a code problem. You need to find every affected repo, open a pull request in each, chase owners and see which have merged. Batch-change and campaign tools exist for exactly that, and they're worth having even if the code itself comes from a simple script.
Single-app upgrades are where IDE agents shine. A developer runs the plan, reads the proposed changes, fixes the build and ships. There's no campaign to manage, so heavyweight platforms are overkill.
Legacy modernisation is its own world. Moving COBOL off a mainframe or re-platforming a VMware estate needs assessment first: what talks to what, what can move as-is and what must be rewritten. Services built for that bring discovery and verification steps a coding agent skips. Data usually travels separately; if your move includes the database, pair the code work with one of these database migration tools instead of hoping the agent handles schemas too.
What drives the bill
The range is wide: free open source engines at the bottom, hosted team tiers starting around four figures a month in the middle, and estate-wide enterprise contracts in the hundreds of thousands at the top. The bill climbs with the number of repos, users and lines of code in scope, and the model matters more than the headline. Open source engines and CLIs cost you engineering time, not licences. Hosted platforms may sell team tiers billed monthly, then custom enterprise plans billed annually. Estate-wide platforms can be quoted as a commercial contract sized to your code estate. Cloud modernisation services may make assessment free and charge for ongoing modernisation work. Campaign features can sit on enterprise plans only, so check which tier includes them. Team tiers sit on a public pricing page, while estate-wide contracts come out of a sales conversation, so go into that call knowing how many repos you'll touch and how many engineer hours the migration would otherwise eat.
A one-hour pilot on a single repository
Pick one real repository that's a version behind, ideally one with decent tests. Give each shortlisted tool the same target, for example a framework major upgrade, and an hour.
Note four things. Did it build afterwards? How many files changed, and were they all ones that needed changing? Would you approve the diff without rewriting it? And could you run the identical change on a second repo and get the same result? That last question separates engines from assistants faster than any sales call. If a tool can't explain what it changed and why, it's not ready for your estate, however clever the demo looked.
Before you scale beyond the pilot, settle the boring questions. Who reviews the recipe or the agent's plan? Who merges in each repo, and what happens to the pull requests nobody claims? How do you roll back a change that passed tests but broke something in staging? Write those answers down first. A migration tool that opens a few hundred pull requests in an afternoon is only useful if your organisation can absorb them, and that's a process question, not a software one.
Do you make AI agents or engines for code migration? Tool makers can ask to be considered for a featured place on this list.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.


