Executive Leadership
CEO Performance Evaluation: Grade the Calls the Agents Couldn’t Make
The board’s annual review of the CEO leans on evidence the CEO produces. When a fleet of agents drafts much of that evidence, directors end up grading the fleet.
I have been CEO of Visiting Media since November 2024, and a lot of what my board sees from me passes through software first. A fleet of specialized agents drafts my operating summaries and chases the follow-ups I would otherwise forget. I also sit on the board of ACT | The App Association, where I am one of the people reading what management sends. From both chairs the same thing bothers me about how boards evaluate a chief executive.
The process assumes the CEO made the evidence.
What the standard guides get right
The pages that rank for CEO performance evaluation agree on the mechanics, and the mechanics are sound. Diligent’s guide walks through appointing one director to run the process, gathering input, delivering feedback and setting next year’s goals, then offers question banks grouped under financial and operating results, strategy execution, leadership and talent, culture and conduct, risk oversight and external relationships. The Ideals guide spends most of its length on committee roles: the compensation committee designs the framework and ties it to pay, while the governance committee compiles director feedback. Several published board policies weight the review half on business objectives and half on leadership competencies such as judgment and decision making.
I would keep all of that. The business-objectives half mostly survives contact with an agent fleet, because revenue is revenue no matter who drafted the forecast. The competencies half is where the trouble lives.
Directors grade what they can see
Think about how a director actually forms a view of a CEO’s communication, or their grip on the business, over twelve months. The board pack. The CEO letter at the front of it. How fast and how precisely the CEO answers a director’s email on a Sunday. Whether the pre-read for the strategy offsite was sharp. Those impressions carry most of the competency score, and in my case agents had a hand in a good share of them.
Then the director reads the pack with an assistant of their own, which I wrote about in writing the board pack so the AI summary holds. One machine drafts and another summarizes, and a human grades the result as if it were a personal quality of the CEO.
Polish has become cheap. A well-run fleet makes a mediocre CEO look organized for about a year, which is roughly one review cycle, and that is long enough to set a bonus.
Grade the decisions the agents could not make
The competency questions in the guides are fine. What changes is where directors look for the answers. I would point them at records that only exist because a person used judgment, and most of these already sit in a well-governed fleet:
- Overrides. Times the CEO overruled an agent’s recommendation or rewrote its draft before it left the building, and what happened afterward. A CEO with zero overrides in a year is either running a perfect fleet or not reading.
- Stops. Agents the CEO paused or shut down, and the reason in one line. Killing a project you championed is hard for anyone. It is the clearest judgment signal available, and the decommissioning record I describe in retiring an AI agent captures it for free.
- The owner column. Count how many agents in the register still name the CEO as owner, against last year. Rising means the CEO is hoarding operating work in software. This is the same count I argue for in AI agent access reviews and in CEO succession planning for an agent fleet, and it belongs in the review because it measures whether the CEO can hand things off.
- One incident, read in full. Pick the worst agent failure of the year and have the board read the write-up. How the CEO talked to the team and the customer afterward tells you more about leading change than any survey question, and the AI incident playbook lays out what that conversation should contain.
None of this replaces the 360 feedback from direct reports. Keep it. Just know that the reports are now partly rating how good the CEO’s agents are at scheduling them.
Where the business-objectives half gets gamed
I said that half mostly survives. The exception is cost. An operating-expense goal can be hit by moving work onto agents while the hours people spend checking that work stay invisible, buried in salaries the goal never counted. An agent that saves twenty minutes of drafting and adds thirty of cleanup still looks like a win on a cost report.
When a CEO hits a cost target through automation, the compensation committee should ask for the review time alongside the invoice before it pays the bonus. I laid out how to price that in who owns the bill for AI agent costs, and the number should already sit on the quarterly page I describe in the AI page in the board pack. If it isn’t there, that absence is itself worth a line in the review.
The self-assessment
The CEO should write the self-assessment without an agent, because it is the one document in the cycle where the board gets to watch the person think, unassisted, at whatever length and in whatever order they choose.
What I would ask my own board to do
Spend one hour of the evaluation meeting on two decisions from the year that the CEO made against the advice of an agent, a direct report, or both. Have the CEO walk through what they knew at the time. Then look at how it turned out, and be honest about luck.
That hour costs less than the survey round. It also tests the competency every guide lists and almost none of them can observe directly, which is judgment under uncertainty, with the record sitting right there in the override log. A CEO who can’t produce two such decisions in twelve months has, in practice, delegated the job to the fleet and kept the title.
I would rather be graded that way. My agents draft a tidier operating summary than I would on a tired Thursday. I don’t want credit for that. It says nothing about whether I should still be running the company a year from now.
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.
Related Reading
Book • Second Edition (October 2026)
Growing Up Fast: How Agile Organizations Adapt, Innovate, and Thrive in the Age of AI
Get the complete operating playbook for running AI agent fleets, agile executive systems, and board-grade technology governance by Jascha Kaykas-Wolff.
Frequently asked questions
How should a board evaluate a CEO who uses AI agents?+
Keep the standard split between business objectives and leadership competencies, but change the evidence for the competency half. Kaykas-Wolff suggests directors look at the CEO's overrides of agent recommendations, the agents the CEO paused or retired, how many agents still name the CEO as owner, and the write-up of the year's worst agent incident.
Why are AI agents a problem for CEO performance reviews?+
Directors judge competencies such as communication and command of the business from the board pack, the CEO letter and email responses. When agents draft much of that material, and directors summarize it with their own AI tools, the board ends up grading the fleet's polish rather than the CEO's judgment.
Can a CEO hit cost goals by moving work to AI agents?+
Yes, and the review should check how. An operating-expense target can be met while the hours people spend reviewing agent output stay buried in salaries. Kaykas-Wolff recommends the compensation committee ask for that review time alongside the model invoice before paying a cost-based bonus.
Should a CEO use AI to write their self-assessment?+
Kaykas-Wolff argues no. The self-assessment is the one document in the evaluation cycle where the board sees how the CEO thinks without assistance, so it should be written without an agent.