It's the team's first time onboarding, and nobody knows which Agents to create or how to divide the work — the biggest fear is spinning up a pile of toys nobody ends up using.
Don't create Agents first. Let an onboarding Agent map the business: how the roles split, which step burns the most people, which systems hold the data. Once it understands, it proposes a design for a hybrid human-plus-Agent organization; after you make the call, new Agents are created one by one through an approval flow, and veteran Agents run onboarding training for the new ones.
Presses for business details, reads real files, researches industry playbooks online, drafts the org proposal plus approval cards and workflow notes for new Agents.
Inventories the user-profile, review, and order-signal databases and flags which market data links are missing.
A newly created support Agent, trained by @guide before starting, with its boundaries written into long-term memory.
A newly created research Agent that settles the data split with @user-insights on the spot before starting work.
We're designing a hybrid human-plus-Agent organization. Rules:
Ops drops real business files into #onboard; @guide reads them and keeps asking where things get stuck the most.
@user-insights inventories the internal databases and reports each market's data picture and gaps.
@guide researches industry playbooks online and maps each one against your current state as a gap analysis.
The two Agents draft org proposals from the ops view and the user view; you pick one to produce the final version.
Approval cards get drafted per the plan and humans approve creation; veteran Agents run onboarding training for the new ones.
Before each new Agent starts, the relevant Agent hands over data interfaces and working templates.
Which Agent has which data tool installed and who needs to query by proxy — stated clearly on a regular basis.
After the pilot site runs for a while, revisit and revise the org proposal.
Pick one site and one product for a 2-3 week pilot with acceptance metrics, then replicate to other sites.
Turn the data-link gaps found during mapping (some markets have almost no user data) into a dedicated backfill project.
Let the Agent push back: a good org proposal should include designs it explicitly recommends against, such as cloning N identical Agent sets per country.