The Benchmark

Kiro can migrate your PHP. Scrutinize the eight phases first.

A repeatable migration engine sounds reliable until you ask what repeatable actually costs at scale.

AIBy The Beacon desk, Lighthouse Strategy

Data pulse

8

Phases in Kiro's PHP-to-Java migration engine

Source: AWS Machine Learning Blog

AWS published a walkthrough this week describing how Kiro turns a PHP-to-Java migration into an eight-phase engine with a mechanical discovery scanner at the front end. The pitch is that a one-time leap of faith becomes a measurable, repeatable process. That word, repeatable, is doing a lot of work in that sentence. Before your engineering team books a sprint, it is worth asking what repeatable actually means when the process involves an AI agent making structural inferences about legacy code it has never seen.

What the eight-phase structure actually tells you

Eight phases signals discipline. It signals that someone thought carefully about sequencing, handoffs, and rollback points. That is genuinely useful scaffolding compared to the more common approach, which is roughly 'give the model your codebase and see what survives.' The mechanical discovery scanner in phase one is probably the most load-bearing component. If the scanner misclassifies a module's dependencies early, that error propagates forward through every subsequent phase. The system is repeatable in structure. It is not immune to compounding misclassification.

This is a calibrated observation, not a dismissal. Structured migration processes reduce variance. Reduced variance is valuable. But variance reduction is not the same as accuracy, and your ops team needs to treat those two things as separate variables when evaluating the output.

The vendor lock-in question you should ask before phase one

Kiro sits inside the AWS ecosystem. The migration engine is built to run there, which means your output artifacts, your inferred dependency maps, and your translated Java modules are all shaped by decisions made inside that environment. That is not automatically a problem. It is a constraint worth naming before you start, not after phase six.

The more pointed question for VPs of Commerce and E-commerce Directors is whether your PHP infrastructure is the actual bottleneck to your commerce roadmap, or whether the bottleneck is something adjacent. Migrations are expensive in engineering attention even when the tooling is sound. Token cost for running an AI agent through a large legacy codebase is not trivial. Latency inside the discovery and translation phases will vary depending on codebase size, dependency complexity, and how many edge cases your legacy PHP contains. None of those numbers are published in the walkthrough. You will need to run your own calibration.

Where your brand sits in the benchmark

Most commerce brands running PHP at this stage fall into one of three rough categories. The first group has PHP infrastructure that is genuinely limiting, blocking integrations with modern commerce tooling, creating latency in checkout flows, or preventing the data connections that agentic commerce requires. For that group, a structured migration engine is worth serious evaluation. The second group has PHP that is old but functional, stable but not growing. Migration for that group is probably a distraction from higher-leverage work. The third group, and this is the most common in mid-market commerce, has mixed infrastructure where PHP touches some systems and not others. For that group, a full migration engine is likely oversized. A targeted rewrite of the specific modules creating friction is almost certainly faster and cheaper.

Best-in-class engineering teams in commerce are not necessarily the ones running the most modern stacks. They are the ones who can accurately diagnose which layer of their stack is creating commercial drag and address only that layer. Kiro's eight-phase engine is a blunt instrument applied to a precise problem. That mismatch is not fatal, but it is worth stress-testing before your team commits the sprint budget.

Three questions to pressure-test

First, can you name the specific PHP module that is creating measurable latency or integration failure in your current commerce stack? If you cannot name it precisely, you are not ready to evaluate a migration engine. Second, has your team run the discovery scanner against a subset of your codebase and reviewed what it classifies as a dependency versus an assumption? The difference between those two things is where migrations stall. Third, what is your fallback if the translated Java modules in phases five through eight produce outputs your QA team cannot verify against your original business logic? If the answer is 'revert,' then your repeatable process has a very expensive exit condition baked into it.

One honest uncertainty: the walkthrough describes the engine but does not publish benchmark data on translation accuracy rates for real-world PHP codebases. That number would change this analysis considerably. If Kiro's eval results on complex legacy PHP reach 90 percent or higher on business logic preservation, the risk calculus shifts toward adoption. Until that data is public and independently verified, treat this as a promising scaffold, not a proven process.

From the publisher

Reading is free. Acting on it is where we come in.

When a story is about your category, a Discovery call turns it into a plan.