FlowsBook a callSign inCreate free accountSign up
AI employees and automation

AI agents that manage other agents: one manager, a team of specialists

How a manager AI employee hands parts of a question to your other AI employees, waits for what each finds and answers in one place, and the rules that keep it in check.

Part of the guide AI employees

By the xMap team · · 4 min read

AI agents that manage other agents let you ask one question and have it split among specialists. In xMap Flows a manager is an AI agent whose Give work to setting lists up to eight of your other AI employees. Each one becomes a tool the manager can call, such as Ask the Road watcher or Ask the Opening watcher: it runs that employee's flow with what the manager gives it, brings back what it found, and the manager writes one answer from all of it.

This is for teams that already run several AI employees and find themselves opening five threads to answer one question. It explains when a manager helps, how to set one up, what it may and may not do, and how to read what it did.

When a manager is worth having

A manager earns its place when the answer needs more than one kind of finding. What should our drivers know before tomorrow's deliveries? needs road closures, the weather and maybe new customers on the route. Is this neighbourhood worth a branch? needs competitors, footfall and who lives there. One flow could do all of it, but you probably already have an employee for each, tuned over weeks. The manager reuses them instead of copying their blocks.

  • Specialists stay simple. Each employee keeps doing one job well, on its own timetable.
  • The manager decides who to ask. For some questions it calls one employee, for others four, depending on what the question needs.
  • Fixes happen once. When you improve the Road watcher, every manager that asks it gets the better answer.

A worked example (made up)

A courier company in Nairobi runs three AI employees: a Road watcher that finds closures and roadworks with Traffic incidents, a Weather watcher built on Weather forecast, and an Opening watcher that finds new businesses along its routes with Nearby places. The operations lead wants one message at 17:00 for the dispatch team.

  1. Add a Schedule for every weekday at 17:00, Africa/Nairobi time.
  2. Add an AI agent after it and set How it works to Work through a big task, since the task comes from its instructions, not from a message.
  3. Write the instructions: prepare tomorrow's dispatch note: closures on our main corridors, rain that will slow riders, new businesses on the routes worth a sales visit; under 200 words, by area.
  4. In Give work to, pick the Road watcher, the Weather watcher and the Opening watcher.
  5. Add an Email to the dispatch team with the report in its message.

At 17:00 the manager plans its steps, asks the Road watcher about the corridors, asks the Weather watcher about tomorrow, decides it needs the Opening watcher for only two of the five areas, and writes the note. If the Weather watcher finds no rain, the note says so in one line and moves on.

What the manager may and may not do

  • No sending while it thinks. If an employee normally emails or posts to Slack, it does not send while the manager is working. In a test run nothing is sent; in a real run that sends, its sending happens once, after the manager has finished.
  • Credits are shared. Each employee's run uses its own credits within what the manager's run has left, and appears in its own run list.
  • Some employees cannot be asked. One that is archived or paused, one private to someone else, or one with a Wait or an Approval in it is refused, and the manager is told why.
  • It sees only your organization. It can only give work to employees of the same organization that its owner may see.

Reading what it did

Open the run and press See how it answered on the agent's line. For a big task you get the plan as a checklist, each step with the employees it asked, what it gave them and what came back, its notes between steps, and why it stopped. On the Team chart, a manages line runs from the manager to each employee it can call, so anyone can see who reports to whom.

Mistakes to avoid

  • A manager over employees that do not work yet. Get each specialist right on its own first. A manager cannot fix a weak employee; it only repeats what it says.
  • Too many employees. Eight is the limit, and fewer is usually better. Each extra one is another thing the manager must decide whether to ask.
  • Instructions without a shape for the answer. Say how long, in what order and for whom. Otherwise the manager's report wanders.
  • Forgetting the budget. Give the manager a monthly budget like any employee, since it pays for the runs it starts.

Where to start

If you do not have specialists yet, the daily operations briefing for every hub template gives you one flow that does several checks, and a weekly brief from your AI employees shows a lighter alternative: the Team findings block reads what all your employees found, without running them again. Pick the manager when you want each specialist asked live, with the question of the day. More on running a team of AI employees is in the AI employees guide, and couriers will find the logistics and delivery page useful.

Questions

How many AI employees can one manager give work to?

Up to eight. Each appears as a tool the manager can call when a question needs it.

Will an employee send emails twice when a manager asks it?

No. While the manager thinks, nothing is sent. In a real run, an employee's sending happens once, after the manager has finished.

Can a manager answer chat messages too?

Yes. Give work to works whether the agent answers each message or works through a big task, so a chat page can be answered by a manager with specialists behind it.

What is the difference from Team findings?

Team findings reads what your employees already found and counts it. A manager runs the employees again with a question of its own and writes from fresh results.

Templates that do this

Read next

More on ai employees and automation · Every article · Templates

Put it to work

An AI employee in xMap Flows does this on live data, on its own timetable. Start from a template or describe it in plain words.