Agentic engineering

A developer describes an intent, a coding agent edits the codebase, tests run, and the developer reviews and ships. The application at the end is an ordinary Laravel or Symfony project. What changes is the loop around it — faster, more deployments, and dependent on how much of the platform an agent can read without a human translating.

# Agents read what the platform publishes

An agent needs to know how deployment works or why a build failed, and it finds out by reading. Documentation wrapped in a JavaScript application behind a search box is close to unusable for that; documentation served as plain markdown at a predictable URL is not.

Every docs article here is available as raw markdown, plus llms.txt and llms-full.txt for retrieval. The open source agent skills go further than reading: /fortrabbit deploy, db down, and ssh run deployments, database syncs, and remote commands over plain SSH and rsync. Logs from PHP, Apache, and jobs can be tailed over SSH, so an agent can quote the failing line rather than asking somebody to look.

# More environments, shorter lives

When a change takes 20 minutes instead of two days, branches multiply, and each one wants somewhere to run that isn't a laptop. An environment maps to a Git branch, books its own components, and is billed pro-rated by the day — one that exists for an afternoon costs a day, and deleting it stops the billing that day.

The deployment trigger is set per environment. The agent's branch deploys on every push; production is switched to manual deployment, so nothing goes live without a person clicking deploy.