The AI Manager
The routing layer that sends every task to the right agent with the right context already loaded.
What it does
The AI Manager is the agent that manages the other agents. Work arrives, and it decides what kind of task this is, which agent owns it, what context that agent needs, and what happens if the agent cannot finish.
It reads from your company brain before it hands the task over. That means the correct account history, the current policy, and the current goals are attached to the task itself. The agent does not have to know to ask. The context is already there.
It also handles the parts people forget. Order of operations. Retries when a system is down. Escalation to a person when the case matches an exception in your playbook. Timeouts, so nothing sits in a queue forever with no owner.
Every routing decision is logged. You can look at any completed task and see which agent touched it, what context it had, and why it went that way. That log is what makes the system reviewable instead of mysterious.
What breaks without it
Without a manager, context loading becomes a manual habit. Someone has to remember to paste in the deal notes, the policy, the last conversation. Most of the time they do not.
That is the failure people describe as the agent forgetting. It did not forget. Nobody gave it the information. The output looks confident and is wrong, and there is no record of what it was working from.
You also get collisions. Two agents pick up the same task. A task with no clear owner sits untouched. An exception that should have gone to a person goes to an agent that answers anyway.
Where it shows up in your operation
- An inbound request arrives with no clear category. The manager classifies it, assigns the owning agent, and attaches the account record before work starts.
- A task fails a validation step. The manager retries once, then routes the case to the named human owner from your exception playbook instead of shipping a guess.
- A quarterly policy changes. The manager loads the current version on every task from that day forward, with no reliance on anyone remembering the change.
How it connects
The AI Manager is the layer between your work and your Agent Teams. It pulls what each task needs from the Company Brain and hands the agent a task that is already grounded. It sends finished work to the AI Judge for scoring, and it routes anything that fails back for a second pass or to a person. The Improvement Manager uses the manager's logs to find where routing itself is wrong, then tightens the rules.