Inigo AnguloContact
Work01How I think02Lab03Writing04
Writing

From ticket to production in five hours: a release process for a one person team

Inigo AnguloNovember 4, 20266 min readAI agent teams

Here is one change on Acervo, as recorded by the ticket, the pull request, the CI run and the deployment. Not a highlight: a Tuesday.

  1. 14:35, ticket writtenA numbered ticket with a measurable acceptance criterion and valid metadata. Without both, the lead returns it before starting.
  2. Branch from mainOne branch per ticket, from an up to date main, never from another ticket's branch. The worker gets its own worktree and database.
  3. 16:56, pull request openedBy the lead, after integrating the worker's output and the reviewer's list of faults. The PR carries screenshots and the preview URL with its commit.
  4. Fourteen checksFive required by the branch ruleset (lint, types, tests, auth templates, secrets) plus the browser suite and the house checks. All read on the exact commit to be merged, not an earlier one.
  5. 17:20, pen test PASSThe security agent runs the full test against local and staging and writes a report with a verdict for that commit.
  6. MergeThe lead merges, only with every required check green on that commit. It does not merge on anyone's say so, including mine.
  7. 19:50, promotionA human command names the commit. The workflow holds the credentials; no person and no agent does. Migrations run before the deploy and are compatible with the previous code by rule: add first, remove in a later release.
  8. 19:54, servedThe final check fetches the production domain and looks for the content. A deployment that reports READY can serve a 404; the first promotion did exactly that, so the check reads the page, not the status.

Why it is five hours and not five minutes

Because three of the steps are human and two are slow on purpose. The ticket takes thought. The merge waits for every check on the exact commit. The promotion waits for a security report. Everything between those points is agents and scripts, and that is where the hours went from days to minutes. The human gates did not get faster; they got cheaper, which is what lets them stay.

Why it is routine

Because nothing in it depends on memory. The ticket has a schema. The decision format is fixed. The checks are the same fourteen. The promotion is one command with one input. A process that depends on someone remembering a step is a process that skips the step on the day it matters, and with agents producing 445 merges in seven weeks, that day comes weekly.

What I would copy first

The whole promotion flow with its lanes is in the Acervo case, chapter 06. The series: AI agent teams for software development and the two human gates. To set this up for your team: AI build and deployment.

Kit

Copy the checklist: Release gate checklist, with a prompt to use it and its changelog.

Working on something like this?

This is the method behind how I build: AI native development with human gates. Discuss a product idea → Or email me.