Morph by Modelcode (modelcode.ai) is one of the AI tools for legacy code migration that you can put through a simple test: run it on a repository as large as yours, then ask four questions. Does it show a plan before any code exists? Can your engineers review each change? Can it show that behavior survived the move? Where is your code executed?
The two-minute video above makes the case with pictures. Below, each question gets the detail you need to ask it well, followed by the answer Morph gives.
Why a demo can't settle the choice
Shortlists for a legacy migration usually mix two kinds of tools: coding agents and assistants, such as Claude Code and GitHub Copilot, and modernization platforms (AWS Transform, Moderne, Morph and others). Nearly every one of them looks capable in a demo, because a demo runs on a repository small enough to fit in one person's head.
Size is the variable a demo hides. In RepoMod-Bench, a benchmark that Modelcode's research team published on arXiv, four agent configurations that the authors describe as state of the art migrated 21 real-world repositories in 8 programming languages, scored by tests the agents never got to see. Their average pass rate was 91.3% on projects under 10,000 lines of code and 15.3% on projects above 50,000. The authors conclude that autonomous modernization at scale remains an open challenge; they do not conclude that agents are poor tools. For a buyer, the practical lesson is narrower: judge every candidate on code of your own size.
Question 1: does it show you a plan before writing any code?
What to ask for: a written plan you can edit, covering the scope, the target stack, the migration approach and your constraints, plus a guarantee that no code is generated until someone on your team signs off. A plan you can read early is the cheapest place to catch a wrong assumption.
Morph's answer: once you submit your modernization goals, it generates a Project Spec from its analysis of your repositories and opens it for review. Modelcode's documentation gives reviewers four checks: whether the destination technologies are right, whether the planned transformation makes sense, whether every requirement is captured, and whether the right parts of the codebase are in scope. You change the spec through a chat that edits it in place, an auto-review compares the new version with what Morph learned from your code and with the goals you first stated, and each of its recommendations must be accepted or rejected before you can approve. The migration does not begin until that approval is given.
Question 2: can your team review everything it ships?
What to ask for: work delivered in increments small enough for a person to read, through the pull request process your team already uses, with the price of each increment known before it runs. One huge diff at the end is a migration nobody truly reviewed.
Morph's answer: the work is split into milestones, and each one reaches your team as a pull request inside your usual code review. Before a milestone starts, a confirmation dialog shows what it will cost in dollars and credits and whether it fits your monthly allotment; it runs only once you agree. Changes sit on isolated branches, and merging stays a decision for your team.
Question 3: how does it prove the migration kept the old behavior?
What to ask for: evidence, gathered by running both systems, that the new one does what the old one did. A passing build and a passing test suite say the code works; they do not say it works the same way. Ask what gets compared, at what level of detail, and what happens when a comparison fails.
Morph's answer: its validation runs the original application and the migrated one and compares their real outputs. For an API, that means status codes and response bodies; for a command-line tool, exit codes and what it prints to stdout and stderr; for a web interface, scripted browser flows, plus an agent that explores the application and records screenshots and video. The tests come from observed behavior, so nobody has to write them by hand. This full comparison is the High validation level, which Modelcode describes as the slowest and most thorough of its three settings. Morph ships the integration tests it used to verify each milestone along with the code.
Question 4: where does your code run during the migration?
What to ask for: the vendor's security documentation first, then a plain answer on where your code is checked out, built and executed, and on which side opens each network connection.
Morph's answer: Modelcode's public trust center shows its compliance with ISO/IEC 27001:2022 and with SOC 2 Type 2, and you can request access to the security documentation there. When code has to stay inside your network, ModelDaemon, the part of Morph that you host yourself, does the checkout, build, test and run steps on machines you control. Modelcode never connects in: the daemon opens its connections itself, outbound only, to the platform that coordinates the work and to the other services it is configured to use.
How to run the test on your own code
Pick one repository that represents your estate in size and shape, not the smallest one you have. Ask every tool on your list the same four questions, in writing, and ask each vendor to show its answer on that repository instead of describing it. Keep the answers in one document, so the comparison stays fair to everyone, Morph included.
If you want to try Morph this way, Morph Pro is priced at $0 per seat; every month includes 40,000 credits, with $0.01 charged per credit beyond that, and you see a milestone's cost before you approve it.
FAQ
What should I look for in AI tools for legacy code migration?
Four things, tested on a repository the size of yours: a plan you approve before any code is written, work delivered in increments your team can review, proof that the new system acts like the old one, and a clear answer on where your code is executed.
Do AI coding agents work on large legacy codebases?
They do, but size changes the results. In RepoMod-Bench, the four leading agent configurations passed 91.3% of hidden tests on average under 10,000 lines of code, against 15.3% above 50,000. Large migrations call for a plan, increments that people review and behavioral verification around whichever tool writes the code.
Is Morph a replacement for coding agents?
No. Morph is built for migration and modernization projects that need an approved plan, milestones and proof; coding agents remain useful for everyday development. The four questions apply to both.
The meaning of verifiable modernization is covered in Verifiable AI Code Modernization: What It Means and How It Works; refactoring programs that span many repositories, in Automated Enterprise Code Refactoring Platform: What It Should Automate and What It Should Prove.
More on Morph: https://modelcode.ai/?utm_source=substack&utm_medium=blog&utm_campaign=video4_twin

