The sales-ops weekly arrives on Monday morning, the way it has every Monday for the last seven months. The lead-routing rules are clean. The pipeline-health note is sharper than the version your VP of sales used to write. The board-prep slide is already in the shared drive. Nobody on the call wrote it last week. Nobody on the call wrote it the week before.
You ask, half in jest, who actually owns the file.
The room goes quiet for about a beat too long.
That moment is the only entry door for this article. Most coverage of the Zeb Evans ClickUp announcement stopped at the layoff number. The line worth keeping sat three sentences down. “Ironically, the people that automate their jobs with AI will always have a job. They become owners of the AI systems - agent managers.” And the line worth keeping after that one: “The 100x org is actually heavily dependent on people, infinitely more than today.”
That is the structural claim this article extends. The 100x organization is more people-dependent, not less. The people who matter most are the ones who quietly automated their own jobs and became owners of the systems that replaced their old role. Most CEOs do not yet know who those people are inside their own org chart.
This piece is the audit. Four moves, in the order a CEO can run them: identify the agent managers on the payroll today, test for succession risk, fix the title and comp and review triad that no longer describes their work, and put the resulting story in language the board can act on.
Move 1: Find them before they leave
The diagnostic is not the title. The diagnostic is the system.
Three signals mark a critical-infrastructure person, in any function. First, their queue evaporated. Tickets that used to back up in their inbox now resolve before they touch them. Second, nobody escalates to them anymore, because the artifact they used to produce arrives on its own. Third, the system breaks when they take a week off. Not slowly, not abstractly. The reports go stale. The hand-offs miss. The customer that always got their early-Monday email gets nothing.
Run the scan across five functions first, because that is where most 80 to 200 employee companies will find the highest density. Sales operations. Demand generation and marketing operations. Customer success. Finance and FP&A. People operations.
Engineering is one example, not the field. Treating it as the field is the read that produced almost every coverage piece of the ClickUp announcement and missed the structural move underneath. The Anthropic Economic Index’s March 2026 report shows the cross-functional spread accelerating. The top 10 tasks on Claude.ai fell from 24% of all traffic in November to 19% in February, as usage spread across more occupations. The same Anthropic report finds that about 49% of jobs already have at least a quarter of their tasks performed using Claude. The breadth is structural, not an engineering artifact.
The conversation to have after the scan is short. Ask the line manager one question. If this person leaves next month, what shows up broken? The answer is the audit output. If the manager cannot answer it concretely, the person is not yet on your radar in the way they need to be. If the manager answers it in three sentences and starts looking nervous, you have found one.
Move 2: The thirty-day departure test
You found them. Now run the departure test. If they leave on the first of next month, what specifically continues working on the fifteenth.
The instinct is to ask for documentation. The instinct is wrong, or at least insufficient. The Panopto and YouGov 2018 Workplace Knowledge and Productivity Report, drawn from a survey of more than 1,000 U.S. workers at organizations of 200 or more employees, found that 42% of institutional knowledge is unique to the individual. This knowledge was acquired specifically for the employee’s current role and is not shared by any of their coworkers. That number was the baseline before any of this happened. The agent-manager era does not solve that condition. It concentrates it. The agent manager has built a system whose logic lives in their head, the prompts they tuned, the edge cases they remember, the routing decisions that look arbitrary until you know the three customers they were built around.
Documentation captures the surface. Apprenticeship captures the layer underneath.
The classic HR response to a critical-infrastructure person is to document the role and engineer out the dependency. The agent-manager case is different in kind. The person did not hoard knowledge that could have been captured. They built and now run a new asset that did not exist before they built it. Documenting them does not retire the asset. It only describes its current state to someone who would then have to take ownership of its continued shaping.
The Monday move is a shadow rotation. Pick one other person in the same function. Give them thirty days of paired work with the agent manager, with explicit permission to ask the questions a documentation effort would never surface. Why does this prompt have that exact phrasing? What happened the last time you let the agent run without that guardrail? Which signal do you check first when something feels off? The output is not a runbook. The output is a second person who could keep the system running for a quarter while you decide what to do longer-term.
This is what the harness compounds, not the model looks like at the human layer. The harness is a body of accumulated decisions. The agent manager is the person carrying it.
The risk is not that this person became important. The risk is that the organization has not yet recognized the system they built as an asset that needs ownership, succession, and governance.
Move 3: Rebuild the title, comp, and review triad
The old triad of title, compensation band, and performance review was designed to measure a job that has now been automated. All three break simultaneously for an agent manager, and breaking them one at a time leaves the other two doing the wrong work.
The title problem is the easiest to name and the least urgent to solve. The role does not exist in the standard HR taxonomy. ClickUp’s announcement coined “agent manager.” Other companies will land on different language. Picking a label can wait a quarter. Picking the right comp model cannot.
The closest current compensation signal comes from AI and digital-talent pay data, even though agent managers will not always sit inside formal machine-learning roles. The WTW 2026 Artificial Intelligence and Digital Talent Salary Survey, covering more than 6,000 organizations and 2.7 million employees across 10 countries, found that across all countries studied, median pay for machine learning roles increased on average by 2% for salaries and 6% in total compensation. Long-term incentives are becoming a more important retention lever. Total compensation for these roles, including salary, allowances, and short and long-term incentives, exceeds $170,000 in the US, compared to around $122,000 in Germany and just under $100,000 in the UK.
The point is not that every company should copy AI-engineer pay bands. The point is that base salary is no longer the whole retention story. When total compensation is moving faster than salary, the signal is incentive design: equity refreshers, retention bonuses, and long-term upside for people who own load-bearing systems. Comp the system the person owns, not the title they used to hold. The 80 to 200 person version of the agent-manager comp package is not a salary band raise. It is a structural shift toward incentive-weighted retention for the small set of people who own load-bearing systems.
The performance review breaks the same way. The standard review measures whether the person shipped the deliverable. The deliverable is now produced by a system the person built. Reviewing them on the deliverable reviews the work the system did. The right questions are different. How many people would it take to replicate the output of the system they own. Is the system getting better quarter over quarter. Is the documentation and apprenticeship strong enough that a replacement could be running it in a quarter, not a year. That is the review. It is closer to a product review than a personnel review, because the asset being reviewed is the system, not the person.
Move 4: The board agenda the slide deck does not have
The hardest part of the agent-manager era for a CEO is not the audit. It is the conversation with the board.
The NACD 2025 Public Company Board Practices & Oversight Survey, conducted across U.S. public-company directors, finds that AI is now a routine topic for over 60% of boards, more than double the share in 2023. The same survey notes that more than 62% of directors are setting aside agenda time for full-board AI discussions. And it observes that, despite the time being allocated, few boards have taken steps to integrate AI into governance structures, strategy, or risk monitoring.
Even if your company does not have a public-company board, the governance question is the same: who is accountable for the continuity of the AI systems now carrying real work?
The gap inside that finding is the opening this article points at. Boards have learned to ask about AI strategy, AI risk, AI vendor concentration, and AI compliance. Boards have not yet learned to ask about the people who own the AI systems. The risk register names model providers and prompts and policy gaps. The risk register does not yet name the three people whose departure would break the new operating model.
This is the same architecture as The Boardroom Risk of Confident AI. Boards take the topic seriously. The topic has a piece their current frame leaves implicit. The CEO move is to name the implicit piece on the next agenda.
The name for it is operating-system continuity. The board agenda item, in three lines. First, the company runs on a set of internal AI systems whose continuity depends on a small set of named operators. Second, the audit identified those operators and their functions, by name. Third, the retention, succession, and comp plan for each is on file with the CHRO and reviewed quarterly. That is the slide. It does not exist in most boardrooms yet. The companies that understand this early will put it there first.
What this is not
This article is not a case for premium-locking three people for ten years. The agent manager era will likely standardize. Some of the systems these people built will turn into platforms. Some will turn into job titles that exist in five years the way product manager exists today, the discipline complete with its own community, conferences, and someone playing the role Marty Cagan plays for product management.
The audit is the move that gets you to that horizon with continuity. Most CEOs reading this can run the four moves in one quarter. The 100x organization is the one that knows which of its people quietly became critical infrastructure for the new operating model.
Most companies do not. Yet.
Questions this article gets
What is an agent manager and is the role real?
An agent manager is an employee, in any function, who has built and now owns an AI workflow that produces what their original job description used to produce. The label was coined publicly by Zeb Evans, CEO of ClickUp, in his 2026-05-21 announcement of a 22% workforce reduction. Whether the title sticks is a separate question. The role itself is observable inside most organizations of 80 employees or more today. The cross-functional spread is supported by Anthropic's March 2026 Economic Index finding that about 49% of jobs already have at least a quarter of their tasks performed using Claude, with usage broadening across occupations rather than concentrating in engineering.
Why the focus on sales ops, marketing ops, customer success, and finance rather than engineering?
Engineering is the most-covered case and the least useful one for this article. Most public coverage of agent-manager dynamics is engineering-coded because the productivity data exists for engineering and not yet for other functions. The structural claim, that a person who automated their job becomes owner of the system that replaced it, is function-neutral. It shows up first in functions with high-frequency, high-volume, rule-bound work: routing, scoring, triage, scheduling, drafting. Those are sales ops, demand gen, customer success, finance, and people ops more than they are engineering. A CEO who runs the audit only on engineering misses the larger pattern.
How do I run the audit without spooking the people I am auditing?
The audit is a manager conversation, not an employee survey. The question goes to the line manager, not to the agent manager directly. The question is also a positive one, framed around continuity and retention, not around redundancy. Identifying someone as critical infrastructure is the start of a retention investment, not the start of a documentation effort that would let you replace them. The order matters. Naming the person, then deciding the retention package and the succession plan, then telling them, is the order that produces both a stronger retention outcome and a calmer conversation.
What is the right comp model for an agent manager at an 80 to 200 person company?
At 80 to 200 employees the working model is incentive-weighted, not salary-weighted, and tied to the continuity of the system the person owns. The WTW 2026 Artificial Intelligence and Digital Talent Salary Survey found median pay for machine learning roles grew 6% in total compensation against 2% in base salary, with long-term incentives becoming the more important retention lever. The 80 to 200 person version of that mix is multi-year retention bonuses, equity refreshers, and restricted stock tied to system continuity rather than OKR shipment. The base salary band is the smaller part of the package.
Where on the board agenda does this actually go?
Inside the existing AI-oversight slot, named as a distinct line item. The NACD 2025 Public Company Board Practices & Oversight Survey finds that more than 60% of boards now treat AI as a routine agenda topic, more than double the share in 2023, but that few have integrated AI into governance structures, strategy, or risk monitoring. Operating-system continuity is the integration move. The board agenda gets three lines: the company runs on internal AI systems whose continuity depends on named operators, the audit identified the list, the retention and succession plan for each is on file and reviewed quarterly.