Run a supervised Factory of AI agent teams that releases only what you approve

Module 1Connect the proactive AI agent teams around one asset brief.2 lessons

1. 1Assign the proactive production AI agent teams

   • Explain why a set of single-job agents can let a stale layer reach every size without anyone noticing
   • Assign the Brief and References, Image Production, and Quality and Delivery teams to the olive-oil brief, each with a named human steward
   • Write each team's rules: what it accepts, what it returns, when it stops and whom it escalates to
   • Separate the decisions that stay with people (image approval, permissions, client contact, label sign-off) from the work the teams may do
2. 2Set the handoff requirements

   • Build a shared production record with asset IDs, accepted references, source versions, permission status, review notes, export requirements and a client-permission field that only the producer fills in
   • Build a handoff table that says what each team must pass on and what the receiving team checks before accepting it
   • Choose Dots or Claude Duet and configure it to follow the approved record, returning traceable versions, precise edit requests and an outstanding-defect list
   • Test that a receiving team rejects an attractive packet built on a superseded brief

Module 2Carry a creative change through the asset set.2 lessons

1. 3Run the production handoffs

   • Run a scoped batch through Brief and References, Image Production, and Quality and Delivery using prepared files
   • Trace every returned version back to its asset ID, source version and brief version in the shared record
   • Handle blocked work so that a stalled item is recorded with an owner and does not move downstream
   • Restart the run from the shared record alone and note what the record failed to tell you
2. 4Reconcile a changed reference

   • Bring a revised label and a new photograph into the record as a new accepted source without deleting the old one
   • Identify every affected size and mark its derived layers, such as shadows and reflections, as stale
   • Rebuild the affected sizes and block outdated exports from delivery
   • Confirm the fix by inspecting the images at viewing scale, not only by reading the record

Module 3Approve the process for the next brief.3 lessons

1. 5Refine the proactive AI agent team instructions

   • Analyse the Day 9 defect list, rejections and restart gaps to find which rules failed or were missing
   • Revise reference rules, revision limits, retries and escalation for each team
   • Record each rule change in the shared record with its version, date and the approving steward
   • Test the revised instructions by repeating the restart from the shared record
2. 6Process a second sample brief

   • Set up a second sample brief in the shared record using files you are allowed to use
   • Run the batch through the refined teams and check every output against the delivery matrix
   • Compare the second run with the first and name which defects recurred and which rule changes worked
3. 7Record creative release

   • Prepare a release record in which the creative owner approves, returns or holds each item, with a reason for every return or hold
   • Name the person who may authorize delivery and record that authorization separately from the agents' work
   • Write the record so it answers likely client review questions before they are asked
   • Test that another operator can find the current brief, accepted source, final variants and any blocking decision