Internal automations without the infra

What the Engineering agent actually ships.

• Glue work as agents

  Describe the sync or alert in plain English — it runs on managed infra, no repo, no pager.

• Production signals watched

  Sentry, PagerDuty, and CI monitored continuously — incidents triaged and filed before standup.

• Tickets that write themselves

  Bug reports become Linear issues with logs, suspect commits, and repro steps attached.

1. Step 01

   Describe it once

   'Sync new Stripe customers to the CRM and ping #sales.' That's the whole spec.

2. Step 02

   Dench builds and runs it

   Managed infra with retries, logging, and alerting built in.

3. Step 03

   Watch the first runs

   Every run logged and inspectable — approve before it goes hands-off.

4. Step 04

   It maintains itself

   APIs change, schemas drift — the automation adapts and tells you what changed.

• Triage Sentry errors into Linear

  Logs, suspect commit, owner attached

• Run scheduled jobs without cron infra

  Retries and failure alerts included

• Sync data between internal systems

  Stripe ↔ CRM ↔ warehouse, no Zapier

• Watch CI and flag flaky tests

  Weekly flake report with owners

• Draft runbooks from real incidents

  Postmortems start 80% written

• Answer 'how does X work' from code

  Grounded in the repo, not tribal memory

• Build internal dashboards on request

  From the warehouse, no BI ticket

• Keep API docs current with the code

  Drift flagged on every release