Pick the first job for an agent
How I decide which repetitive job to automate first, and which calls still need a person.
The first agent should take one named job off a person. Not a desk. Not a vision. One leak.
What counts as a job
A job has a start, a finish, and a gate. Catalog QA starts with a live listing and finishes with a proposed fix. Repricing starts with costs and competitor moves and finishes with a price, or an exception.
If you cannot name the gate, you do not have a job yet. The gates I use are spend, outbound messages, and record changes. A person stays on that gate.
What I refuse as a first job
A full desk rebuild. A file you import and hope. Unattended spend. A standing engagement that starts before something is live. Those are different offers.
How the first one ships
Name the job and the gate. Build against the systems you already use. The agent drafts. A person approves. You leave with that job running.
Work already public on this site: catalog QA against a golden record, a daily repricer with margin floors, catalog health scored before a suppression.
If the leak is the desk itself — inbox, follow-up, and handoffs that do not talk — that is a connected desk, not a named job.
Want this run for your operation?
This playbook comes straight from client work. If you'd rather have it done than do it, that's what JS Labs is for.
GET IN TOUCH