COXSWAIN
ROEN

Work · 04

A ticket-to-pull-request pipeline, AI-assisted, with enforced quality gates

An AI reviewer reads every change before a person does, backed by tests and static checks. That cut human code review by about 90%: people review business and product logic, the AI proposes and people decide what goes into the product.

AI in production

90%less human code review: people check business and product logic
Context
Delivery pressure, a small team and the temptation to let AI write code straight into production.
What we did
We built a pipeline where an AI agent takes the ticket, proposes the implementation and opens the pull request. The work is backed by unit and automated tests, PHP checks (custom PHPStan rules, PHPUnit coverage gates) and React front-end checks, and on every change to a pull request an automated AI reviewer requests changes or approves. People review business and product logic. Used daily in delivery, including multi-agent setups.
Result
About 90% less human code review, and faster delivery without lowering quality: the same gates for AI-written and human-written code.
Stack
Claude Code, PHPStan, PHPUnit, React, GitHub Actions

Three layers before a person looks

The pipeline has three layers. Tests: unit tests and automated suites, many of them written with AI help. Checks: custom PHPStan rules and PHPUnit coverage gates on the PHP side, and the React front-end checks. And on every change to a pull request, an automated AI reviewer reads the diff and does what a reviewer does: it requests changes, or it approves.

What people still review

People review business and product logic: does this change do what the customer needs, and is it the right thing to build. Everything else is the job of the tests, the checks and the AI reviewer. That removed about 90% of the review work and moved human attention to the part that needs judgment.

The gates are the same for code written by people and code written with AI. The AI proposes, people decide.

What broke along the way

The rough edge was auto-merge. It produced merge conflicts that broke the AI reviewer, and we had to work around it. It is the kind of problem that only shows up once a pipeline like this runs on real work every day, which is why we start with a pilot and not with a rollout.

Services

The service behind it

Symfony upgrades

Symfony upgrades, module by module, without a rewrite

We upgrade Symfony products one major version at a time, module by module, tests first, while your team keeps shipping. Backed by a Zend Framework to Symfony 7 migration done in 3 months with zero cutover incidents.

 4-6 weeks

AI-assisted delivery pilot

One to three concrete wins: generated and verified tests, AI-assisted PR review, a module upgraded or migrated in production, a stabilized front end or a smaller AWS/Azure bill.

 Test generationModule migrationCost optimization

Work

More work

Migration · Enterprise SaaS

From a mixed PHP and Node.js stack to Symfony 7, with behavioral parity

A product used by large companies, moved to a modern foundation piece by piece, without users noticing.

 3 monthsfrom a mixed PHP and Node.js stack to Symfony 7 Integrations · Microsoft 365

Outlook calendar sync at enterprise scale, without loops or notification storms

Outlook calendars and the platform stay correctly in sync for thousands of users, without duplicate notifications.

 2-wayOutlook sync at enterprise scale, without loops or storms Cloud · Kubernetes

Clusters and databases moved between AWS, Azure and Google Cloud, in production

Live production moved from AWS to Azure in two weeks, with about five minutes of downtime.

 5 minof downtime moving live production from AWS to Azure Security

Hardening a multi-tenant platform: isolation, uploads, SSO, API keys

When a large customer asked how secure the platform is, the answer was backed by code, not by promises.

 SSO + 2FAhardened across the platform

All work

Contact

Let's talk

Tell us what you have, in your own words. You do not need to know which technology the system runs on; we will work that out. You get back questions and a written proposal.

Write to usor directly at [email protected]