Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.productbrain.com/llms.txt

Use this file to discover all available pages before exploring further.

Jobs are not tasks. They are acceptance criteria — observable proof that a bet is being delivered.
If you’re used to Jira, Linear, or kanban boards: forget that model here. There is no backlog → doing → QA → done pipeline. A job is either proven or it isn’t.

​
What makes a good job?

A job label describes what you’d see if the approach is working. The description says where to verify it. Implementation detail goes in the notes field.
GoodBad
Label”Price comparison shows for scanned barcode""Add API call to fetch prices”
Label”Upload progress bar reaches 100% for 500MB file""Implement chunked upload”
Label”Share link shows read-only tree""Create share endpoint”
The good labels are observable results. The bad labels are implementation steps — those belong in the notes field, where builder agents read them.

​
Three fields, three audiences

  • Label = what you can observe (strategist checks this)
  • Description = where to verify it — UI, devtools, API response
  • Notes = how to build it — implementation detail for coding agents

​
Maturity

The quality bar for this job, set at planning time:
MaturityMeaning
MVPIt works, it’s testable, rough edges acceptable. The approach can be validated.
ReleasablePolished enough to stay. Not embarrassed when a stranger uses it.
Maturity is not completion status. Use status: "done" for that.

​
Status

A job is either done or not done. There are no intermediate states.
  • done — the checkpoint has been verified
  • Absent — work is not complete

​
Granularity

  • Each job should be verifiable in under a minute. If you can’t confirm it quickly, it’s too big — split it.
  • Don’t create jobs until the approach is actively being worked. The approach captures the bet. Jobs are the checkpoints when you’re ready to execute.
  • The approach is sometimes the job. For simple approaches, don’t decompose at all. Only create jobs when there’s genuine complexity worth tracking separately.

​
Execution discipline

When working through jobs — whether you or an AI agent:
  1. Do the work for this checkpoint
  2. Verify it passes (check what the description says to check)
  3. Mark it done
  4. Move to the next job
Never batch multiple checkpoints. Each job is a gate. If you skip verification, you compound errors.