Executive Leadership
CEO Succession Planning When the CEO Runs the AI Agent Fleet
Every board I know has an emergency succession plan. It names an interim CEO, says who calls the lenders, and sits in a folder until somebody needs it. The ones I have read say nothing about the agents.
That gap used to be harmless. A CEO’s working knowledge lived in their head, their inbox and the people around them, and an interim leader could rebuild most of it in a few weeks of hallway conversations. Some of mine now lives somewhere else.
I run a fleet of specialized agents across my own operating work as CEO of Visiting Media. It started as automation I built for myself, for work I was tired of doing by hand, and it grew into something with a register, a delegation matrix and a named owner for every agent. A lot of those owner fields say my name. I also sit on the board of ACT | The App Association, so I read succession plans from the other chair too, and I would not want to be the director who learns about a CEO’s private fleet on the morning the CEO is unreachable.
What the succession guides cover
The pages that rank for CEO succession planning cover the same ground with different amounts of data. They open on how high CEO turnover has run, and on how confident directors say they are anyway. They recommend an annual review of the plan, a bench of ready-now internal candidates, and a separate emergency protocol for the sudden departure. The newer ones add that boards should weigh AI fluency when they write the profile of the next CEO.
I agree with every one of those points. They are all about the person.
None of them ask what the outgoing CEO was personally running, and that is the angle I care about here. If the CEO built or owns agents that do real work, those agents are a key-person dependency in the most literal sense, and an interim CEO inherits them whether or not anyone wrote them down.
What actually breaks on day one
Picture the CEO gone without notice on a Tuesday, for any of the ordinary reasons boards plan for. The interim takes the chair on Wednesday. Here is what I would expect to go wrong in a fleet like mine, roughly in order of how much it would hurt.
- Nobody can turn anything off. The agents keep running on schedule, and the credentials that let someone pause them sit with the one person who is not there. This is the one that keeps me honest.
- The delegation matrix loses its signer. My delegation of authority for agents ranks actions by how reversible they are, and any increase in what an agent may do without a human gets signed by the CEO personally. With the CEO gone, the matrix freezes. Frozen is fine for a week. It is a problem by the next quarter-end.
- Outputs lose their reader. Some agents produce work that only the CEO acted on. A briefing nobody opens is wasted compute. A briefing nobody opens that also flags a contract renewal is a missed renewal.
- The weekly graded sample stops. Mildly bad, at first.
The first item carries most of the weight. I have already lived a small version of it: an agent I replaced in the spring was still waking up on its old schedule in late summer, holding a token nobody had revoked, and I was the person who had set it up. That was me forgetting my own system with the CEO fully present. An interim with no context would have a much harder time finding that kind of thing at all.
An agent only one person can stop
If only one person knows how to turn an agent off, the company does not really own it.
The fleet handover file
My fix is a short document that lives inside the emergency succession plan, kept by the general counsel or the corporate secretary next to the bank signatories and the board contact sheet. It holds the fleet register (every agent, its owner, what it can access, where it runs), and for each agent a named second person who can pause it. It records where the credentials live and who besides the CEO can rotate them. That count should never be zero. (The count of agents running on personal credentials should be, a line I already argue for in an AI risk appetite statement.) It also carries the current delegation matrix with a note on who signs changes while the CEO seat is held by an interim.
Write it so the interim can read it in ten minutes. Most of the fleet should be safe to leave running, and the file should say so plainly, because an anxious interim who pauses everything on day one will break things the company depends on (the boring reconciliation jobs, mostly).
The second-person column is where this gets uncomfortable. Filling it in forces the CEO to admit how many agents exist because the CEO likes them, with no one else in the company who would notice if they stopped. Those are retirement candidates. I have retired more agents than I launched this year, and filling in that column is one of the faster ways I know to find the next few.
Test it with a quiet week
Paper plans fail in ways the paper does not show. The test I would ask a board to run is simple and a little awkward to request, which is part of why it works. The CEO takes a week off and does not touch the fleet at all. No logins, no checking the dashboard from a phone at the airport. Whoever is named in the handover file runs it. Afterward that person writes a page on what they could not do, and the CEO does not get to edit it.
The fleet will not fall over in a week. Something will drift, though. A token will turn out to live in a config file only the CEO could open, or a scheduled job will fail and alert a phone that is in a drawer. Each of those is a line the handover file was missing.
Run it once a year, at the same time the board reviews the rest of the succession plan, so it becomes part of a cycle that already exists instead of a new ritual someone has to remember.
The incoming CEO should not inherit it whole
The handover file solves the emergency. A planned succession has a different problem. Agents a CEO built for their own work are tuned to that CEO’s judgment, their reading habits, even their tolerance for being interrupted at night, and a successor who keeps them unchanged is running somebody else’s operating system.
I would give the new CEO the register in their first month and the authority to retire anything on it, then ask them to re-sign the delegation matrix in their own name before the first board meeting they chair. Re-signing sounds ceremonial, and partly is. It is also the moment the new CEO reads every line and discovers which ones they would never have agreed to. Treat it like retiring an agent, applied to a whole fleet at once, with the consumers found first and the keys confirmed dead afterward.
What directors should ask
Add one question to the next succession review: which agents would stop being supervised if the CEO left tomorrow, and who would pause them? If management cannot answer from a document, the plan has a hole, and it is cheaper to find it in a committee meeting than during a crisis.
This belongs with the nominating and governance committee, which already owns the plan, rather than with whoever handles technology. The question also fits naturally beside the ownership and concentration items in what boards should ask about AI, and it is the kind of thing a technology board advisor can raise without anyone hearing it as an accusation. Boards are right to put AI fluency into the profile of the next CEO. They should also find out how much of the current CEO’s job already runs without them, because that part leaves the building on the same day they do.
This article is part of the Executive Leadership cluster, focused on board governance and the operating discipline required to run AI systems responsibly at the executive level.