Executive Leadership
Executive Presence When AI Agents Run the Work
An agent’s log tells me what it did. It almost never tells me what anybody did with the result, and executive presence starts in that handoff.
An agent’s log tells me what it did. It almost never tells me what anybody did with the result. That is the first test of executive presence around AI: I stay with the work after the model call and tool call end, because the consequential decision often sits with the person who receives the output.
I want the audit trail to continue through that handoff. It should show who reviewed the output, what they changed, who approved it, and who relied on it. A clean agent run can still produce a bad operating result. I cannot understand accountability by reading only the portion of the record generated by the agent.
Put Review Time on the Same Bill
The model invoice is the part finance can see. The time somebody spends checking the agent’s work does not appear beside it. When I evaluate an agent, I put human review time on the same bill as tokens, routing, and repeated calls. Otherwise, a cheap run can hide an expensive operating process.
That bill needs an owner. If a team owns the agent’s output, it also owns the cost of checking, correcting, and approving that output. Executive presence shows up in the willingness to carry the full number into an operating review, especially when the token line looks good and the human line does not.
Make Every Approval Gate Explicit
I name each approval gate, the person responsible for it, and the decision that person is making. The name matters because the audit trail needs to show where the agent stopped and human authority resumed. An unnamed gate leaves the handoff unclear, even when the workflow records an approval.
The gate also belongs in the audit trail. I want to know which output reached it, what the reviewer saw, whether anything changed, and who released the work. This gives me a record that connects agent activity to human judgment. Without that connection, an approval label says very little about how the decision happened.
Report the Fleet Without Polishing It
At the board level, I use one page with the same rows every quarter. It covers the agents running the work, the ones we turned off, their full cost including review time, and the approval structure around them. Consistency matters because directors can see movement instead of getting a fresh collection of pilots and vendor logos.
I also include an honest line about which parts of the board pack an agent drafted. That disclosure connects the reporting process to the controls we claim to use elsewhere. Executive presence with AI is visible in the record: what ran, what stopped, what people checked, and who accepted the result. The quality of that record becomes part of how the board reads management.
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
Frequently asked questions
What does executive presence look like when AI agents run work?+
It looks like visible ownership of the full operating chain. I expect leaders to follow the audit trail beyond the agent, price human review time into the bill, assign each approval gate, and report the fleet to the board with the same rows each quarter.
Why should an audit trail continue after an agent stops?+
Because the consequential action often happens after the final model or tool call. The continuing trail should record who reviewed the output, what changed, who approved it, and who relied on it. That is how I connect agent behavior to the human decision and its operating result.
How should a CEO calculate the cost of an AI agent?+
Count the token bill and the human time spent reviewing, correcting, and approving the work. I assign that combined cost to the team that owns the agent’s output. A low model invoice says little when the time required to check the work remains outside the accounting.
What should an approval gate record?+
It should record the output presented, the named reviewer, any changes, the decision made, and the person who released the work. I treat the gate as part of the audit trail, because an approval label without the underlying handoff does not explain how judgment was applied.
What should CEOs report about AI agents to the board?+
Use one page with the same rows every quarter. Include the agents running, the agents turned off, full cost with review time, approval ownership, and an honest line about which parts of the board pack an agent drafted. Repetition lets directors compare the operation over time.