04 · AI
AI in your development process
Everyone is talking about AI. We work with it, every day and for every client. Agents read through logs and code, review every pull request, investigate tickets before anyone starts on them and build on their own where they can. It makes us faster, and it makes the work cheaper for our clients.
What we use ourselves, we set up for your team as well. Not a chatbot beside the work, but agents inside your existing Jira, Git and CI pipeline, following the same coding rules as your developers, with limits enforced in code rather than left to good intentions.
How we use it →- 01
Incident analysis
An agent reads slow logs, server metrics and application logs alongside the code. What used to take days of searching now often surfaces the cause within the hour: one query, one line of config.
- 02
Code review
Every pull request reviewed against the project's own coding standards, with inline comments. Clean work moves on to test; a blocker goes back to the developer with an explanation.
- 03
Ticket preparation and estimation
Before the team discusses a ticket, it has already been investigated in the real code: root cause, files, pitfalls, plus a reasoned story point estimate. Open questions have already gone to the right person.
- 04
Autonomous development
A label on a ticket is enough: the agent creates the branch, writes the code, runs the checks and opens the pull requests. QA rejections get fixed on the same branch.
- 05
Upgrades and takeovers
Modules compared against their originals, changelogs and patches read, dead code pointed out. Inherited codebases get mapped sooner, and upgrades go through faster.
- 06
Guidelines and guardrails
One set of coding guidelines per repository that Claude, Copilot, Cursor and Junie all read, and hard limits on what an agent may push and merge. Fast, but never unsupervised.
HOW WE USE IT OURSELVES
Agents work for every client, from a store on a single server to an Adobe Commerce platform with three storefronts. On our largest development team, five routines run on their own every working day, from the morning prep to the release email. The first ticket built fully autonomously was ready within eleven minutes and passed QA and code review first time.
