Managing a team with AI
When part of your team is agents, who does what?
Anyone who puts agents to work on a team finds out that management is still there, just somewhere else. Every task still has three roles: who decides (always a person), who executes (a person or an agent) and who checks before it ships.
Who decides
Always a person. On a sales proposal, the partner decides price and scope.
Who executes
A person or an agent. The agent can put together the monthly report from the numbers.
Who checks before it ships
Someone reads the reply to the customer before it goes out.
When a role has no name
The agent decides something nobody knew about, or the work reaches the customer without anyone having looked at it. A short rule helps: the agent can execute without asking, but it doesn't decide without someone knowing.
How I solved this in my own work
I built Arroway because of this. The team's decisions live in one place that every AI reads before acting, and what an AI proposes only counts once a person approves it.
Frequently asked questions
- Who does what when part of the team is agents?
- Every task has three roles: who decides (always a person), who executes (a person or an agent) and who checks before it ships.
- Can an AI agent decide on its own?
- It can execute without asking, but it doesn't decide without someone knowing.
- What happens when one of the roles has no name?
- The agent decides something nobody knew about, or the work reaches the customer without anyone having looked at it.
- Who checks what the agent delivers?
- Someone on the team, before it ships. On a reply to a customer, for example, a person reads it before it goes out.