_edited.png){kind=small}
Watch Me Run an AI Team: How I Coordinate the Work
Your research is ready. The email still needs a banner. The video is edited, but nobody has checked whether the upload is finished. You open another chat to move things along—and realize you have become the person every task waits for.
When I run an AI team, I give specialists defined responsibilities and use a coordinating assistant to keep their work connected. An AI team is a group of software agents—assistants that can work through assigned tasks—organized around a shared business goal, with clear instructions, tools and handoffs. The owner still makes the decisions that need human judgment.
In my video, “Watch Me Run an AI Team,” I show how I organize that work around Eva, my ChatGPT dot. What changed for me was the way I organized the jobs. More disconnected chats would have left me chasing the same unfinished work in more places.
If you want help building this around your own business, Zero-Employee Entrepreneur teaches the wider process: choose a mission, define the specialists and decide how work moves between them. Start here with the part you can apply today.
In This Article
How I run an AI team around one coordinator
Eva is my dot—the assistant I use to coordinate the work. I think of her role as a chief of staff: someone who can look across assignments, identify what is waiting and bring the next decision back to me. The specialists have narrower jobs.
Maven: email marketing strategy.
Rainmaker: content research.
Pulse: preparing, uploading and scheduling email campaigns within the assigned scope.
Jax: video editing.
Pixel: video animations.
Atlas: YouTube uploads and scheduling when authorized.
Amy: community management.
Theo: helping create, investigate and improve the AI employees themselves.
These names belong to my setup. You do not need to recreate every role. If your bottleneck is researching useful topics, an entire video-production team may be the wrong first investment of your attention.
Think of a small production studio. The producer tracks the whole job; the editor focuses on the cut; the researcher supplies the evidence. Asking the producer to personally do every task would make the specialist roles pointless. My AI chief of staff guide goes deeper into the coordinating role.
OpenAI describes dots as assistants that continue work between conversations. My example is a particular working arrangement, not a promise that every account has identical features or access. Check what your account supports before designing the process around it.
Follow one job through the team
Here is how the email workflow fits together. Eva asks Maven to research ten email ideas based on past winners. I approve the direction. Eva then sends the work to Pulse to prepare the campaign, including the copy and banner, and handle the authorized scheduling work.
The approval matters because choosing the message is a business decision. A technically complete email can still sell the wrong idea. A handoff should carry the decision already made, not force the next specialist to guess it.
Video follows a similar pattern. Rainmaker surfaces ideas. I approve the direction and write the script. Eva coordinates Jax’s editing and Pixel’s animations, then Atlas handles the YouTube stage. Before moving on, the team needs to know which version is approved and what “finished” means for that particular job.
For your first workflow, use a short handoff brief:
Approved input: the exact topic, source document or asset to use.
Next owner: the specialist responsible for the next result.
Deliverable: what must come back, and where it should be saved.
Check: the standard that proves the result is ready.
Boundary: what can proceed and what needs your decision.
If you want the foundation behind that sequence, my guide to AI workflows explains how inputs, steps and outcomes fit together. The practical test is simple: could the next specialist begin without asking you to reconstruct the entire conversation?
Give your first specialist seven essentials
The tempting move is to name an employee and immediately hand it a broad responsibility: “Handle my marketing.” That leaves the agent to guess your priorities, standards and limits. A name helps you organize the work; the operating instructions make the role useful.
These are the seven setup points I cover in the video:
A definition of done. Describe a result you can inspect. “Prepare three email concepts with evidence and a recommendation” is easier to check than “improve my marketing.”
A trigger. Say what starts the work: a direct assignment, a coordinator’s handoff or a supported recurring schedule.
Specific instructions. Explain the sequence, quality standard and corrections you expect it to remember.
Context. Supply the relevant business facts, source files and examples. Make the approved version obvious.
Tools. Give the specialist the supported access it needs for the assigned task. Instructions alone do not connect your accounts.
Skills. Save a repeatable method built from examples, feedback and testing. In a supported setup, recording a process can help capture steps that are easy to omit.
A written operating guide. Record the role, boundaries and handoffs somewhere the agent can read and reuse.
I use Markdown files for that last piece. Markdown is plain text with simple formatting such as headings and bullets. Think of it as a readable job manual, not programming you must learn before you can begin.
My seven-part AI delegation brief helps you turn a broad request into an inspectable assignment. For a task you already perform well, the next step is to capture the process as an operating procedure, then test whether the specialist can follow it.
Start with preparation work you can review. Ask for a draft, comparison or proposed plan. Once the first result meets your standard, decide whether to expand the responsibility. Do not confuse a connected tool with permission to publish, spend or send on your behalf.
Keep the mission visible
A team can produce a lot of work without moving the business forward. You might end the week with more posts, more research and a beautifully organized folder while the offer that needs attention remains untouched.
That is why I separate responsibility for watching the mission. The mission includes the goals and targets the team exists to support. Its job is to help review whether the work contributes to those goals, rather than rewarding activity simply because it is easy to count.
For a first test, choose one bottleneck and one observable result. Perhaps the result is a reviewed campaign ready for your approval by Thursday. Track where it stalled, how much correction it needed and whether the completed work served the goal. Broader sales or lead results need their own evidence; faster output alone does not prove revenue growth.
Give every recurring job a return address: the business goal it is meant to serve. If you cannot name that goal, pause before adding another specialist. My AI project management guide explains how to make ownership and progress visible across the work.
Start with the job that keeps coming back to you
You may look at my team and feel as if you need a weekend to build the whole thing before it becomes useful. You don’t. Look at the assignment you keep restarting, the file you keep passing between tools, or the correction you have made three times this week.
That is a better starting point than copying every employee’s name. In my own setup, I spend much of the initial work clarifying instructions. Theo helps me investigate and improve employees when something breaks. That ongoing repair is part of building a dependable team; it is not evidence that you picked the wrong name or need another subscription.
My view is that the first useful AI employee should give you back the attention needed to build the second. Pick a bounded responsibility, explain the finish line and inspect the first result. Save the correction so the next run has a better starting point.
That is the approach I teach in Zero-Employee Entrepreneur. If you are ready to build your own team around a meaningful business goal, start with that first role. You can decide what the wider team needs once you have seen real work move.
Questions about running an AI team
Do I need to know how to code?
You can begin by defining a role, providing instructions and files, and reviewing its work. The complexity depends on the task and tools. Start with capabilities your existing setup supports rather than assuming every workflow needs custom software.
Do I need ChatGPT dots to organize specialists?
Dots are the coordinator in my demonstrated setup. You can still organize responsibilities, keep instructions together and carry handoffs yourself while testing the process. Automatic coordination depends on the capabilities and access available in your chosen environment.
What if I cannot see dots in my account?
OpenAI says access is rolling out gradually, with eligibility varying by plan and region. Check the current dots access guidance before purchasing a plan for this feature. Local-computer work also requires the connected computer to be available.
How much does an AI team cost?
There is no single price for this setup. Cost depends on subscriptions, task usage and connected services. Set a budget for one small workflow first and review its actual usage before expanding the team.
Can I let the team publish everything automatically?
Set boundaries for each role and task. In the workflow I describe, I approve the creative direction; any publishing, sending or scheduling work must stay within the authority you granted. Ask for reviewable drafts while you establish the standard.
Comments