Agents are already doing part of the work. Nobody sat down and decided which part. So people redo what the agent did because they don’t trust it, the agent does things nobody reviews, and when something goes wrong there is no one whose name is on it. The managers end up refereeing every case by hand.
What we’ll do
- Put the real work on the table. Three to five recurring workflows that matter, as they actually run, not the tidy version.
- Sort each part. Human-owned. AI-supported, a person decides. AI-executed, a named person reviews. Not delegated, because the relationship or the consequence needs a human.
- Test the human claims. Anything the team wants to keep human has to pass four questions: who answers for it, who it affects, whether it needs judgment, and whether the people affected can push back. This is where fear and turf get separated from judgment.
- Name who decides. For everything an agent touches, one person who owns the outcome. For everything the team can’t settle, a written question for the level above.
- Say it out loud. The team drafts the sentence it would say to the rest of the company about what stays human here, and why.
What you leave with
One page: the team’s handoff map, with a name next to every part the agents touch. A short list of decisions that need to go up. And a sentence the team can say to a customer, a new hire, or itself about where the line is.
Who this is for
- A leadership team of a company where agents are already in the workflow.
- One working team and its manager, as a pilot before the rest of the company.
- Heads of People or Operations who are being asked to write the AI guidelines and want them to come from the work, not from a template.
What this is not: tool training, a prompt library, or a governance policy written by outsiders.
How it continues
Most teams run this once, then keep it alive two ways: a monthly ninety-minute practice where managers bring the cases that didn’t fit the map, and 1:1 coaching for the people whose roles moved most. Both are on the organizations page.
The details
- Format: Facilitated working session, whole team in the room
- Length: Half a day
- Where: Your office in the Bay Area, or online
- Group: One leadership team, or one working team with its manager
- Sample it first: the public session, What Should Stay Human?, runs the individual version of this in San Francisco
Your facilitator
Nima Imani is an ICF-certified, ontologically trained coach with more than ten years in software, data and AI, including EY and Neo4j. He runs the weekly founder coaching group at SF Commons in San Francisco and founded InsightsOut.
About InsightsOut
InsightsOut (insightsout.work) is a San Francisco practice founded by Nima Imani, working on the human side of AI change and leadership. For organizations: workshops and coaching for leaders, managers, and teams on decisions, trust, and changing roles, and on what should stay human when agents join the team. For founders: coaching on conflict, overwhelm, and decisions. For everyone: public workshops in San Francisco and research built from what people share in those rooms.