AWS has published a practical guide to using its Kiro AI coding environment to migrate PHP applications to Java, with AWS Transform offering a route to run the same method across multiple applications. The central idea is to make tests, human approval and the original application’s behaviour the referee, rather than trusting generated code to look right.
AWS AI Watch analysis
What happened
The guide sets out an eight-phase workflow, starting with a scanner that inventories PHP models and flags ambiguous cases for human review. Before Java code is generated, teams build a parity harness to compare the original application’s responses with the port. The remaining phases cover scaffolding, translating data and application logic, and checking parity before cutover.
AWS says the same versioned skill files can drive Kiro interactively, one application at a time, or AWS Transform for parallel portfolio migrations. The guide’s sample is deliberately limited: it ports four of 37 models, and its parity harness was not built. AWS calls that the sample’s biggest gap, a useful admission in a field where demos sometimes glide past the awkward bits.
Why it matters
The method gives developers concrete checkpoints: scan before translating, review flagged ambiguities, establish a working comparison test, and approve stages before proceeding. Teams can also reuse improvements to the migration rules across projects, and, according to AWS, across both Kiro and AWS Transform.
That makes this more than a promise that an AI agent can rewrite an application. It is a proposed process for making a migration repeatable and checking whether the result behaves like the software it replaces. The sample is not evidence that the approach has been proven across large production migrations.
Our read
The strongest idea here is the parity harness, not the number of files an agent can change before lunch. If a migration tool must show where the new application differs from the old one, developers have a better basis for catching costly surprises. Start with a small, testable slice, and do not treat the sample as proof that a full migration is ready to run unattended.
What to watch
- Whether AWS publishes results from migrations that use the parity harness on complete applications.
- How well the scanner identifies framework-specific models and flags ambiguous cases.
- Which review and approval checkpoints AWS Transform supports when running migrations at portfolio scale.
Discussion spark: Would you trust an AI-assisted migration more with a strict parity harness, or with human approval at every phase, and which should be non-negotiable?
Sources and evidence
- Build a repeatable PHP-to-Java migration engine with Kiro (29 October 2026, 18:14 UTC)
not affiliated with or endorsed by Amazon Web Services (AWS)