Subagents for big upgrades: scouts, workers and a reviewer
One agent context is not enough for a large TYPO3 multi-site upgrade. How I split the work into read-only scouts, workers with disjoint file sets and an adversarial reviewer.
Senior engineering with AI agents: what changed in my day
The agent does the legwork now. Scoping, reviewing and saying no became most of my job as a senior TYPO3 developer. Where my time goes, and what juniors should still learn by hand.
Rector and Fractor for TYPO3 v14: a workflow that stays reviewable
One big Rector run over an extension gives a diff nobody can review. My workflow: one target version, one set per run, one commit per set, and a clear list of manual work.
A verification gate for AI-written PHP: lint, PHPStan, tests
AI-written PHP looks plausible, which is the problem. One Composer script runs lint, PHP-CS-Fixer, PHPStan and PHPUnit, and a Stop hook keeps the agent working while it is red.