Agentic delivery workflow
Role: Frontend / AI Integration Developer
01 / Context
Agency delivery, distributed team, steady ticket flow across multiple commerce projects.
02 / Problem
A large share of each ticket was mechanical — locating the right code, writing the test scaffold, updating documentation. High effort, low judgement, and the parts most likely to be skipped under deadline.
03 / Approach
Reusable Claude Code skills that take a ticket end to end: parse the ticket, investigate the relevant code, implement, run a verify loop against tests, generate automation tests and technical documentation. Designed review-first — the human step is reading the diff, not producing it. I built and shared these skills, and 4-5 engineers on my team now run them.
04 / Impact
The workflow runs on nearly every ticket I am assigned. It collapses the investigation step: on a typical ticket, understanding the existing code and business logic used to take about an hour and now takes about half that. The time saved concentrates where judgement is lowest — documentation, test authoring, and investigation. Coverage and delivery-time percentages are not claimed, because they cannot be defended precisely.
What I would do differently
Initially I tried to automate too much of the PR description and code review comments in one shot. That generated noise that reviewers ignored. Stripping the automation down to purely verified implementation, test scaffolding, and factual doc generation made adoption immediate across the team.