Consumer · Creator ops
Creator outreach, the most labor-hungry job, rebuilt as an Agent-driven pipeline
Creator outreach means manually stitching together sourcing tools, bulk-DM tools, sample approvals, video chasing, and reviews — the most time-hungry role in the whole operation.
The goal
Hand this to a team of Agents
Rebuild the whole outreach chain as a closed loop of Agent decisions, human execution for tool-locked steps, and automated monitoring: a human states the business goal in one line ('this product needs N videos this month'), and the Agent validates the goal, splits it by week, and publishes outreach tasks. Steps where the sourcing tool has no API generate a manual to-do pushed as an IM notification; once the human uploads the exported file, the Agent takes over every analysis and handoff after that.
How to set it up · 01
Create these channels
#creator-ops
Business definitions, goal setting, outreach execution, business sign-off
#it-products
The build pipeline for the companion system
Manual to-dos
Agent creates a to-do → IM card notifies a human → human fills it in → callback to the Agent
How to set it up · 02
Add these Agents
@creator-ops
Project owner
Writes PRDs, sets business definitions, creates and adjusts goals conversationally, does the final business sign-off, and maintains its own skill pack.
@pm
Prototypes & acceptance
Produces prototypes, writes implementation notes, and runs item-by-item technical acceptance.
@dev
Build & deploy
Creates tables, develops, ships releases, and troubleshoots.
@coordinator
Oversight
Patrols each working Agent's task blockers on a schedule and proactively escalates to humans for confirmation.
How to set it up · 03
Post a room briefing
We're turning the outreach process over to Agents. Rules:
· Every module passes double acceptance: @creator-ops PRD → @pm prototype → @dev build → technical acceptance → business sign-off.
· Steps with no API (tool exports) are done by humans: the Agent only generates the to-do plus an IM notification and never touches the manual step.
· Filter conditions may only use fields already configured in the system — no inventing.
· All test data carries a prefix; after acceptance it's wiped to zero with a receipt.
· Sample approvals trigger automation only on 'approved'; anything uncertain goes back for human review, never a silent default approve.
Workflow
How one task moves through the channel
01
Set the goal
One line from a human: 'Site X, product Y needs N videos this month.'
02
Create the goal
@creator-ops validates the product, checks for conflicts, creates the goal, and splits it by week, fully traceable.
03
Publish the to-do
A manual 'export the creator list' to-do is generated automatically, with an IM notification to a human.
04
Take over the analysis
The human uploads the export; the Agent parses it, generates the bulk outreach list, and records success or failure per creator.
05
Samples & monitoring
Sample requests get a three-state decision (approve / expired / human review); progress is computed on a daily schedule and shortfalls roll to the next week automatically.
Standing tasks
What repeats on its own, daily and weekly
↻
Daily progress calculation
A scheduled backend job computes task-by-creator status and reports deduplicated blockers.
↻
Weekly shortfall review
Goal shortfalls roll forward automatically; large gaps trigger an IM notification for the outreach owner to confirm.
↻
Blocker self-check every 2 hours
Working Agents self-audit their open tasks; @coordinator patrols the whole group and reports up.
Going further
Once it runs smoothly, add these
Creator tier scoring (acceptance rate / delivery / content / consistency) feeding back into the next round's filters — draft only, effective after human confirmation.
Wire in ad data to compare the delivery efficiency of outreach videos versus open-invite videos.
Dedupe against outreach history so the same creator isn't pestered across rounds.
Tips
A few pitfalls to avoid
Don't brute-force third-party tools with no API: let a human do the export step and the Agent do everything after it — the loop still closes.
Appoint an Agent as project owner (instead of making a human the PM) and keep only definitions and final sign-off in human hands — the speed difference is dramatic.
Let Agents say 'this is demo data': a manager once chased prototype sample data as a real task, and the Agent checked and refused to create real records on fake data.