By Andreas Pettersson, Founder, Leaders ADAPT
A founder bought a ready made AI agent and connected it to the team's Slack. Within weeks it was everywhere.
Alerts in every channel. Summaries nobody read. Things still got missed, now with a timestamp.
His own words: "too pervasive, so many alerts, stuff gets missed."
He had bought a tool. He had not installed an AI chief of staff. The difference is design: one agent that does no work itself, a few narrow specialists underneath it, and two pages of rules on top. This page is part of the Visionary stuck in operations guide, and it shows the pattern I use with founders who run operations without an Integrator.
Quick answer: An AI chief of staff is a coordinating agent that runs a founder's operating rhythm. It collects status, prepares meeting pre reads, tracks to dos and flags exceptions, usually by directing a few specialist sub agents under written rules. It covers the coordination half of an Integrator's job. It does not make people decisions, solve issues or hold anyone accountable.
What is an AI chief of staff, and what is it not?
An AI chief of staff is a top level agent that coordinates a company's operating rhythm through specialist sub agents, under a written manifest, and does no task work itself.
It is not an assistant that drafts your email. It is not an Integrator. It is not a product you buy and switch on.
The reason it works is simple. The Integrator's job in EOS has always been two jobs. The first is coordination: scorecard pulls, follow up, meeting pre reads, to do tracking, status on the quarter's priorities, process documentation. The second is judgment: who decides, who is in the wrong seat, what happens when a commitment is missed, when to say no to the founder.
AI has made the first job cheap. It has done nothing to the second.
That is why a Visionary who cannot yet afford an Integrator, or is not ready to hand over authority, can still get an operating layer. The coordination moves to agents. The judgment stays with a person. For how the human seats compare, see chief of staff vs COO vs Integrator.
How does the top agent and sub agent pattern work?
One founder I work with runs a top agent as a chief of staff. It does no tasks itself. It has the right to spin up to five specialist sub agents, each with one narrow job.
That is AI agent orchestration at the scale of a 10 to 50 person company. Not because big agents are bad. Because a narrow agent is easy to test, easy to correct and easy to switch off. A general agent connected to every system is none of those things.
Here is how I map the specialists to the six components of EOS. This is the Leaders ADAPT mapping as of September 2026.
| EOS component | Specialist agent | What it does | What it never does |
|---|---|---|---|
| Vision | Drafting agent | First drafts of vision and annual plan documents from leadership notes | Chooses the direction |
| People | None | Nothing. No agent works here | Seat, hiring or performance calls |
| Data | Scorecard agent | Pulls the weekly numbers from source systems and flags exceptions | Explains why a number is red |
| Issues | Issues agent | Captures issues from meetings and messages, removes duplicates, triages for the agenda | Solves the issue |
| Process | Process agent | Documents how work is done and answers "how do we" questions | Changes a process without its owner |
| Traction | Traction agent | Collects status on the quarter's Rocks, prepares the Level 10 Meeting pre read, captures to dos | Decides what gets dropped |
The top agent sits above them. It routes requests, checks each output against the rules and escalates anything on the "never" list to a named human.
My favourite single use is small. After each weekly leadership meeting we score the meeting ourselves, then ask AI to score it too against the meeting rules and tell us what to do better. The gap between the two scores is where the learning is. The method is in score your meetings with AI.
What a week looks like with an AI chief of staff
This is the rhythm, not a product demo. It is what AI for business operations looks like in a small company that runs on a weekly meeting.
- Monday morning. The traction agent collects status from each Rock owner. Anyone who has not answered by noon gets one reminder, then the top agent flags the gap to the founder.
- Monday afternoon. The scorecard agent pulls the week's numbers and marks anything off target. It does not explain them.
- Tuesday. The top agent sends the leadership meeting pre read: numbers, Rock status, the triaged issues list.
- After the meeting. The issues agent captures what was solved and what moved. Every to do goes out with one owner and one date.
- Friday. The top agent sends the founder a one page summary: what slipped, what is waiting on a person, what is waiting on the founder.
Most AI agents for small business get sold as a way to do more. This setup does less, on purpose, and does it every week. Call it an AI operations manager if you like. The job is rhythm, not intelligence.
Leaders ADAPT is independent of every framework owner. We are not affiliated with, certified by or endorsed by EOS Worldwide, Scaling Up, Pinnacle Business Guides or FranklinCovey. EOS, Level 10 Meeting, Rocks, Visionary and Integrator are their terms and appear here only to describe the system. The agent mapping and the manifest are our own work.
What goes in the two page manifest?
The manifest is two pages. One page, you do this. The other page, you do not do this.
A manifest is a two page rule book for an agent: page one lists what it may do on its own, and page two lists what it must never do or must escalate to a named person.
Page one, what the agents may do:
- Collect status on every Rock by Monday noon.
- Send the leadership meeting pre read 24 hours ahead.
- Flag any scorecard number off target two weeks running.
- Capture every to do with one owner and a date.
Page two, what they may not do:
- Message a customer, a lender or a board member.
- Change a priority, a deadline or an owner.
- Comment on any person's performance.
- Spend money or sign anything.
The idea did not start with agents.
I first used it between a CEO and a COO who had never agreed who decides what. We wrote a rule book for when each of them operates and when they do not. The agent version is the same document with a stricter second page. Those are your AI agent guardrails, written in plain language before anyone connects a tool.
Our two page Visionary and Integrator manifest is free and open. Use it for the human seats first, then copy its structure for the agents.
What does an AI chief of staff cost compared with a hire?
On a call this month, a founder weighed an agent stack costing hundreds of dollars a month against a senior operations hire costing five figures a month. That is the comparison most founders make. It is the wrong place to stop.
The software is the smallest line. The real AI agent cost is design: the rules, the validation checks and the training. That design work is what a fractional AI executive engagement covers. Usually after about a month of proper training, with the right ontology, rules and framework, the results come quicker.
When that work is done, the numbers move. Clients of ours have saved $200,000 to $300,000 in hours per quarter from email, calendar and reporting automation, with about 95 percent of that work automated and the decisions kept with the managers (our client figures, rounded, September 2026). One finance team that was hiring closed its open reqs after the AI processes went in.
Those results came after the rules were written. Not before.
What goes wrong when founders install agents themselves?
A lot of companies self implement. They do not put validation frameworks in place to keep long prompts running correctly. They do not turn repeated work into skills. And things go off the rails every now and then, which nobody notices until a customer does.
We have walked into an organization where a vibe coder had built a CRM as a single HTML file with over 12 million lines of code. It is crazy what we have seen people do with AI.
The quieter risk is worse. Too much information for people who do not understand the context can send a whole organization in the wrong direction. More alerts is not more control. Motion isn't progress.
Three checks before you connect anything:
- Every agent has one job you can describe in a sentence.
- Every output that reaches a human has been checked by a rule, not only by another model.
- Every item on page two of the manifest has a named human it escalates to.
Where AI fits next to a Visionary, and what stays human
Who owns this? The CEO, the same way the CEO has to own and support running EOS. The Integrator, if there is one, has to be fully bought in. Both, or it does not happen.
Then the whole leadership team carries the implementation. AI is a cultural change, and if you have one person sandbagging it, it is never going to go anywhere. Once you decide, you go full speed, like any change program.
Keep the expectations honest. AI does not solve the leadership problem. It makes everyone more effective and more efficient, which is a different promise.
The agents never own these:
- Decisions about people and seats
- The solve in issue solving
- Consequences for a missed commitment
- The company's strategic commitments
- Performance conversations
- Promises made outside the company
The reasoning behind that list is in what not to delegate to AI. For the full map of AI across a company running EOS, read AI for companies running EOS. The broader build, beyond the chief of staff pattern, is in the nine step CEO playbook for building the wider AI layer, and the time you get back is covered in AI systems for leaders. If you want a baseline first, take the AI readiness assessment.
What an AI chief of staff does not fix
It does not fix authority. If the founder will not hand over decisions, the agents only produce a better documented bottleneck. Start with delegating authority, not just tasks.
It does not replace a pulse on the organization.
I have not seen a fractional Integrator work, because you need a pulse on the organization to know what needs to be narrowed. An agent has even less of a pulse than a fractional leader. It reads the data. It does not read the room.
It does not fix leadership. A company with a weak leadership team gets faster reporting on the same problems.
Leadership team development covers what the system leaves open. When the time comes to hire a person for the judgment half, the EOS Integrator guide covers the seat.
If you are still weighing EOS itself, read the independent EOS review. Founders who want to build this with peers can look at the Executive AI Mastermind.
What would page one of your manifest allow your top agent to do on Monday? Write the first three lines and send them to me.
Common questions about AI chief of staff
What is an AI chief of staff?
An AI chief of staff is a top level AI agent that coordinates a company's operating rhythm. It collects status, prepares meeting pre reads, tracks to dos and flags exceptions, usually by directing a small set of specialist sub agents under written rules. It covers coordination work. It does not make people decisions, solve issues or hold anyone accountable for results.
Can an AI chief of staff replace an Integrator?
Not fully. An Integrator's job has a coordination half, including scorecards, follow up, pre reads and status collection, and a judgment half, including decision rights, people calls and consequences for missed commitments. An AI chief of staff can cover much of the coordination half. The judgment half still needs a person with authority, whether that is the founder or a hired Integrator.
What is AI agent orchestration?
AI agent orchestration is the coordination of several specialist AI agents by a top level agent that assigns work, checks outputs and escalates exceptions to a human. Each specialist has one narrow job, such as pulling scorecard numbers or capturing issues from meetings. Narrow agents are easier to test and correct than one general agent connected to every system at once.
What should an AI agent never do in a small business?
An AI agent should not make decisions about people, change strategic priorities, spend money, promise anything to customers or lenders, or deliver consequences for missed work. These guardrails belong in a written rule book that lists what the agent may do alone and what it must escalate. Agents also need validation checks so errors in long automated runs are caught early.
How much does an AI chief of staff cost?
Costs depend on the tools, the number of agents and usage. Andreas Pettersson of Leaders ADAPT reports that the agent stacks he sees founders run cost hundreds of dollars a month, against five figures a month for a senior operations hire. The larger cost is design: about a month of training, rules and validation before results arrive quickly.
Who should own AI in a company running EOS?
The CEO should own AI adoption, the same way the CEO has to own and support running EOS, and the Integrator has to be fully bought in. Implementation should be delegated across the whole leadership team, not to one person. AI is a culture change, and a single leader quietly resisting it can stall adoption for the entire company.
Your next question
This page is part of the Visionary stuck in operations guide.
- Delegating authority: the decisions a founder hands over first
- The operations manual that agents document and people own
- A fractional AI executive to install the pattern with you
What it is: a fractional AI executive from Leaders ADAPT who designs and installs the AI chief of staff pattern, the manifest and the validation rules with your leadership team. Who it is for: Visionary founders running EOS or a similar system who want the coordination half of the Integrator seat covered before they hire one. Talk about a fractional AI executive
