● Use cases

Start from a ready workflow. Shape it to your team.

Six ready workflows cover the work most teams repeat: interface design, features, pull request reviews, a broken build on main, alerts and upkeep. Each one is a starting point, not a template to adopt. We shape it with you: which sources feed it, which steps it runs, where a person decides, which checks must pass and what it may spend.

01 Design task

Agents build the screens. You approve the draft and the result.

Our flagship workflow, and the one we run on our own product. Agents draft the screens in your Storybook, check them and fix them before anyone is asked. You give the idea a draft review and the finished screen a final review.

  • Brief checked before work starts
  • Up to two agent fix rounds
  • A preview link for every version
Every stage, step by step →
UP TO 2 FIX ROUNDS BRIEF PLANNER AGENT DRAFT STORYBOOK REVIEW CHECKS · AGENT DRAFT REVIEW THE IDEA POLISH WIRED · TESTED FINAL REVIEW THE FINISHED SCREEN MERGE ON ITS OWN MAIN

02 Tracker issue

Issues in your tracker become merged changes

Issues that are clear enough to build still wait for someone’s free afternoon. Label one in your tracker, or assign it to the agent: it is planned, built, reviewed and merged through the queue. A person approves the merge where your rules ask for one.

  • Issues from your tracker
  • Review fixes capped per change
  • Merges on its own, or after your approval
  • Required checks pass on the exact commit
FIXES · CAPPED OR MERGES ON ITS OWN ISSUE LINEAR · JIRA PLAN PLANNER AGENT BUILD CODING AGENT REVIEW CHECKS · AGENT YOU APPROVE WHERE YOU REQUIRE IT MERGE MERGE QUEUE MAIN

03 Pull request

Every pull request gets a first review

Pull requests from people and other tools queue up for a first read. A review agent reads each one and posts its review on the pull request: review comments and a check run. If the branch moves while it works, the review is refused instead of landing on code that changed.

  • Pull requests from people or tools
  • Review comments and a check run
  • Refused when the branch moved
PULL REQUEST FROM PEOPLE OR TOOLS REVIEW AGENT READS THE CHANGE HEAD CHECK DID THE BRANCH MOVE? REVIEW POSTED COMMENTS · CHECK RUN REFUSED BRANCH MOVED

04 Main CI fails

A broken build on main starts its own repair

The build on main breaks, and every branch behind it inherits the failure. The factory opens a repair task on its own and an agent works on the fix, which passes the same checks and merge queue as any change. When the build is green again, the repair task resolves.

  • Starts from main’s CI
  • Same checks and merge queue
  • Resolves when the build is green
MAIN BUILD BROKEN CI FAILS GREEN AGAIN REPAIR TASK RESOLVES REPAIR TASK OPENS ON ITS OWN FIX CODING AGENT MERGE CHECKS · QUEUE

05 Alert

An alert becomes an investigation report

An alert fires, and the first hour goes to finding where to look. Any monitoring tool can send it to the factory through a webhook, and an agent investigates and writes a report. It changes no code: the report is the result.

  • Any monitoring, through a webhook
  • Signatures verified, duplicates dropped
  • A delivery log and a dry run
  • Presets for Sentry, Datadog, PagerDuty and Grafana
NO CODE CHANGE SAME ALERT AGAIN ALERT ANY MONITORING WEBHOOK SIGNATURE CHECKED DUPLICATES DROPPED INVESTIGATE AGENT INVESTIGATION REPORT MAIN

06 Calendar rule

Routine upkeep runs on a schedule you set

Dependency bumps and other chores are never urgent, so they slip. A calendar rule starts one run per slot. Runs never overlap, and a slot that is skipped or missed is recorded, not lost. Once you trust it, the workflow merges on its own.

  • Calendar rules, such as every weekday at nine
  • One run per slot, no overlap
  • Missed slots recorded
  • Merges on its own when you allow it
CALENDAR RULE MAIN ONE RUN PER SLOT SKIPPED: PREVIOUS RUN STILL GOING DEPENDENCY BUMP CHECKS

● All use cases

Every use case runs on the same factory

Each of these runs on the same factory as the six above. Each one is a workflow your team picks up and shapes: its sources, steps, checks and the points where a person decides. You can also ask an agent to set one up for you.

  • Person approves A person signs off before the result counts.
  • Only if stuck A person is asked only when the work stops: the budget runs out, checks stay red, or there is no way forward.
  • No decision The result is a review or a report; nothing waits for a person.

These are defaults. Each team sets its own.

Shipping features 6

  • Build a feature from the trackerTracker issue → planned, built, reviewed and mergedOnly if stuck
  • Ship a small fixTracker issue → a merged fixOnly if stuck
  • Urgent hotfixIssue labelled urgent → a fix through the hotfix workflowPerson approves
  • Build from a specAPI call with the spec → a merged change with its testsOnly if stuck
  • Plan before buildingTracker issue → a plan for a person to approve, then the changePerson approves
  • Add an endpoint from a contractTracker issue → route, validation and tests, mergedOnly if stuck

Code quality and upkeep 7

  • Weekly dependency bumpsCalendar rule → one change per slot, checks greenOnly if stuck
  • Remove a retired feature flagAPI call → the flag and its dead code removedOnly if stuck
  • Clear the lint and type backlogCalendar rule → a few small fixes per slot, mergedOnly if stuck
  • Split an oversized moduleTracker issue → split by responsibility, tests unchangedOnly if stuck
  • Delete unused codeCalendar rule → removals, merged once checks passPerson approves
  • Stabilize a noisy test fileTracker issue → the tests fixed and mergedOnly if stuck
  • Keep conventions consistentCalendar rule → small corrective changesOnly if stuck

Migrations and modernization 6

  • Upgrade a framework versionTracker issue → upgraded code, checks greenPerson approves
  • Move off a deprecated libraryTracker issue → the replacement, merged in stepsPerson approves
  • Adopt newer language featuresTracker issue → converted files, tests greenOnly if stuck
  • Migrate files in small batchesCalendar rule → one batch per slot, each its own changeOnly if stuck
  • Containerize a serviceTracker issue → a Dockerfile and compose file, build verifiedPerson approves
  • Add a language to the appTracker issue → strings extracted, translations addedOnly if stuck

Testing 5

  • Raise coverage on a moduleCalendar rule → new tests, mergedOnly if stuck
  • Reproduce a bug with a testTracker issue → a failing test, then the fixOnly if stuck
  • Repair a broken build on mainMain CI fails → a repair change, resolved when the build is greenOnly if stuck
  • Fix a flaky test that broke mainMain CI fails → the test stabilized, or resolved when a rerun is greenOnly if stuck
  • Flag untested codePull request → a review that names the missing testsNo decision

Pull request review 4

  • Review every pull requestPull request → a published review and a check runNo decision
  • Check against team rulesPull request → a review that applies your checklistNo decision
  • Review migrations for riskPull request → a review that flags destructive changesNo decision
  • Review changes from other toolsPull request from a bot or an agent → a review, refused if the branch movedNo decision

Incidents and on-call 4

  • Investigate a production alertAlert through a webhook → a report with the likely cause, no code changeNo decision
  • Turn error reports into tasksWebhook from your error tracker → a task, routed by your intake rulesOnly if stuck
  • Draft a postmortemWebhook when an incident closes → a written timeline and follow-up actionsNo decision
  • Fix what an alert foundAlert through a webhook → a fix, proposed as a changePerson approves

Documentation 3

  • Weekly changelogCalendar rule → a changelog drafted from the week’s merged changesNo decision
  • Keep docs in step with codeTracker issue → the docs changed in the same change as the codeOnly if stuck
  • Document an endpointTracker label → an API reference, mergedOnly if stuck

Design and UI 2

  • Draft a screen for reviewTracker issue → Storybook stories and a review sheet for your draft reviewPerson approves
  • Polish and wire a screenDraft approved → the screen wired into the product, for your final reviewPerson approves

● Shaped to your team

You set what comes in, how it runs and where people decide

Every workflow on this page is a starting point. Change it in the web app, or ask an agent: describe the change in plain words, and it drafts the new version for you to check. Every change is a new version: before a change to intake or routing goes live, you see its effect on the last seven days of incoming work, and rolling back means activating the previous version.

01

What comes in

Which sources feed a workflow, how they rank against each other and which work goes where. Any webhook or API call can start one too.

02

How it runs

The steps and their order, the instructions each agent gets, the environment it runs in and the checks that must pass.

03

Where people decide

Which steps wait for a person, how many review rounds are allowed, what a task may spend and whether merges wait for approval.

Keep your process.
Add the agents.