The Chief Agent Officer Question

TL;DR

Every C-suite role we take for granted (CISO, CDO, CHRO) started as somebody else's side hustle and earned its own seat once the work scaled. The next role on that arc is the Chief Agent Officer: the executive who owns the agentic workforce itself, distinct from the AI strategy behind it. You probably don't need to hire one today, but you do need to start the work now. The companies that wait pay for it twice: once in incidents, once in the recruitment scramble.

A familiar pattern

Every C-suite role we now take for granted started as someone else's side hustle. Operations sat under the founder until it grew big enough to need a COO. People management sat under operations until it grew distinct enough that a Chief People Officer (or CHRO) became the only sensible answer. Information security sat under the CIO until Citicorp appointed Steve Katz as the world's first Chief Information Security Officer in 1995. Data sat under everyone and no one until Capital One created the first Chief Data Officer in 2002. The pattern is consistent: a function emerges, gets absorbed into an existing role, scales past the point where that's tenable, and eventually claims its own seat.

I think we're at that point with the agentic workforce. The AI-strategy version of this story has already started playing out. In March 2024, the US Office of Management and Budget told every CFO Act agency to appoint a Chief Artificial Intelligence Officer within four months (OMB M-24-10): a named executive accountable to the agency head, with explicit responsibility for AI governance, risk, inventory and adoption. The UK followed in January 2025 with the AI Opportunities Action Plan and Matt Clifford's AI Opportunities Unit inside Cabinet Office. Today, every major US federal agency has a CAIO. The private sector is patchier; Lloyds, NatWest, Mastercard, Accenture and Walmart have all appointed variants, though no consistent pattern has emerged.

Almost every one of these roles is about AI strategy: which models to bet on, how to govern data, how to handle ethics and risk. That's a real job, and HBR, BCG and MIT Sloan have argued the case for either a Chief AI Officer or distributed C-suite ownership well enough that I don't need to repeat it here.

None of those roles is about the agentic workforce itself: which work agents should and shouldn't do, how they're hired and retired, what they cost, how their performance is managed, where they sit alongside (and sometimes instead of) people. That's the next discipline waiting to claim a seat, and I think it needs its own C-suite owner: a Chief Agent Officer. The phrase has been floating in analyst commentary and vendor decks for a while; what's missing is a concrete definition of what such a role would actually own.

The right way to think about the CAO is as the agent-side equivalent of the CHRO at the level of discipline, not org chart. Where HR owns the people workforce (strategy, policy, hiring, performance, off-boarding), the CAO owns the agent workforce on the same terms, with the obvious caveat that agents aren't people and the operating model is different. AI strategy is a separate conversation, sitting much closer to data strategy and the CDO: which models to bet on, how to govern training data, what the broader AI direction is. The two are easily conflated; they shouldn't be. Whether your business needs a CAO now, and how you bring the role in without inventing another fiefdom, is what I want to work through here.

The gap nobody actually owns

Pull apart what an agent really is once it's in production. It's a piece of software that holds an identity, gets credentials, makes decisions, costs money to run, can be wrong in ways that hurt customers, and degrades without anyone noticing when the underlying business process changes around it. In other words, it behaves like an employee. A weird, fast, scalable employee, but an employee.

I'm using that analogy deliberately. Agents aren't people, but treating them as digital employees, an emerging category that neither HR nor IT was built to manage, gets you to the right operational questions faster than any of the alternatives. The lifecycle is the same shape even if the ontology isn't.

Now ask who in your business is accountable for that whole lifecycle today. The whole thing, across hire, performance, cost, and retirement.

The CTO owns the platform the agent runs on. The CIO owns the estate it integrates with. The COO owns the operations the agent participates in. The CISO owns the risk surface. The CDO owns the data it eats. The CHRO (or Chief People Officer) owns the humans it sits next to. The CFO owns the budget line, if anyone has bothered to create one. The CAIO, if you have one, owns the strategy that put the agent there in the first place.

None of them own the agent as a worker. The hiring decision (do we build, buy, or rent this one?). The performance review (is it still useful six months in?). The off-boarding (who decommissions it, and how do we know nothing depends on it?). The cost control (we have an estate of agents now, what are they doing, and which of them earn their keep?). The behavioural management (this one is starting to make decisions that look like the previous one we retired for the same reason).

This is the operational layer that AI strategy doesn't touch, and that the rest of the C-suite is structured to ignore. It's where the value of an agent estate is either realised or evaporates without anyone noticing, and right now it falls between the cracks. That's not a hypothetical. I've seen plenty of customers running hundreds if not thousands of agents with no governance framework in place, a meaningful share of them shadow agents that nobody can quite remember commissioning. The gap is already here. The question is what shape of accountability fills it.

The agentic curve is steeper

The CISO took a decade to become standard. Most of the industry looked at Steve Katz in 1995 and said "interesting, but my CIO has it covered." Then Sarbanes-Oxley landed, the FSA tightened operational-risk expectations on UK banks, and breaches started making the front page rather than the back. By 2015, you couldn't run a regulated business in the UK or the US without a CISO; by 2020, you couldn't get cyber insurance without one. The CDO followed a similar arc: roughly fifteen years to become a default role in financial services, another five to spread elsewhere. Both started as luxuries. Both became table stakes. The companies that waited until the inflection point paid for it twice: once in incidents, once in the scramble to recruit at the same time as everyone else.

The agentic equivalent is coming on a faster clock. Gartner expects 33% of enterprise software applications to embed agentic AI by 2028, up from less than 1% in 2024, and 15% of day-to-day work decisions to be made autonomously by then. Marc Benioff is targeting a billion Agentforce agents by the end of 2025. McKinsey reckons agents alone could automate 44% of US work hours. Even if those numbers are half right, the curve is steeper than the one CISOs and CDOs rode in on.

What I think a Chief Agent Officer actually owns

The CAO is a strategy-and-operations role for the agentic workforce. The CHRO owns the people strategy: hiring, performance, policy, off-boarding. The CAO is the same idea, scoped to agents: they own the workforce strategy, set the policies, and run the estate that results.

Seven things, roughly. Click, or tap, to expand each one.

Notice what's not in that list. The platforms agents are built and run on stay with the CTO. Data governance stays with the CDO. Cyber risk stays with the CISO. People policy stays with the CHRO. AI strategy (including model and capability choices) stays with the CAIO if you have one, or distributed across the C-suite if you don't. The CAO is the agent-workforce owner: strategy, policy and operations for the estate, sitting alongside the others rather than absorbing them.

The fair pushback is that none of this is novel as a discipline. FinOps already exists. CMDB already exists. MLOps already exists. Responsible AI policy already exists. Why a new C-suite role rather than a senior director under the CTO with a steering committee?

The answer is that each of those existing disciplines folds cleanly into an existing C-suite domain because what they manage is software, spend, or process. FinOps is about cloud spend; it sits under the CFO or CTO. MLOps is about model deployment; it sits under the CTO. Agents are a new category, somewhere between software and people: they hold identities, make decisions, sit in workflows next to humans, and (per the Gartner and McKinsey numbers above) will become a fast-growing proportion of how work gets done. The seven things in the list above don't fold cleanly into one of the existing domains. That's the test for whether a function deserves its own seat: when distributing it across the existing leaders genuinely stops working. I think we're getting to that point with the agentic workforce, even if today the work can still sit under a senior director with a steering committee around them.

Where the role bumps into others

The cleanest way to think about delineation is by the question each role answers.

The CTO, CIO, CDO and CISO answer the obvious technical questions about platform, estate, data and risk. The four roles where the boundary actually does work are these. The CAIO answers what is our broader AI direction: which models, which capabilities, which data, on what terms? The COO answers how does the business run end-to-end? The CHRO answers how does the people workforce evolve alongside the agent one? The CAO answers which work should we agent at all, who owns each agent, is it earning its keep, and is this still the right call?

Three boundaries matter. CAIO/CAO: the CAIO sets the broader AI strategy (models, capabilities, data, cross-cutting policy); the CAO sets the agent-specific strategy and runs the estate (which work gets agented, where the red lines sit, how agents are hired and retired). Where there's no CAIO, AI strategy sits with the CDO, the CTO, or distributed; the CAO still owns the agent workforce.

COO/CAO: the one most likely to get fudged. The COO owns how the business runs; the CAO owns how the agentic workforce inside it runs. The CAO can sit under the COO at first, exactly as the CISO sat under the CIO for a decade. Scale and consequence eventually trigger the promotion.

CHRO/CAO: a productive tension, provided the two stay separate. The CAO and CHRO will need to co-design role definitions, performance frameworks, even something resembling employment policy for digital workers. Folding agent management into HR is a category error; agents aren't people. Folding it into IT is the older category error; agents aren't applications either.

The counter-argument worth taking seriously is the one HBR has been making about AI strategy: central roles can let everyone else disengage. "AI is the CAO's problem" is exactly the failure mode that produced shadow IT in the 2010s. I haven't fully resolved that here, and I want to flag it rather than wave it away. My current view is to keep distributed accountability for AI outcomes in every business unit, while centralising the operating model for the agent estate. Most companies will need both, and a CAO without an AI strategy lead (or vice versa) is only half a structure. But I can see the world in which centralising agent governance produces exactly the disengagement HBR warns about, and that risk is real enough to name.

How would you know it was working?

Two layers matter for measuring whether this is working: the role itself and the estate it runs.

A CAO is doing the job well when they can answer four questions without notes. Who owns each agent in the business? What is each one costing and producing? Which agents have we retired this quarter, and why? Where did we say no, and why? If those answers exist, the registry, the unit economics, the lifecycle and the "should we" line all exist with them.

A subtler test: the CHRO is free to focus on people, the CTO is free to focus on platforms, the COO is free to focus on operations, because the CAO is holding the agent-workforce threads that would otherwise pull on each of them. If those leaders are still being dragged into agent-specific debates, the CAO function isn't yet doing the work it exists to do.

The agent workforce is in good shape when:

• Every agent in production has a named human owner, and the shadow-agent count is trending toward zero.
• Cost per agent is known, tracked, and trending in the direction the business intends.
• Time agents save is showing up as new work the business cares about, rather than quietly absorbed into "we're just less busy now."
• Drift is caught in quarterly review cycles, before customers feel it.
• Retirement is a planned event with a runbook.
• Humans working alongside agents say their work got more interesting, with engagement scores to back that up.
• The board can read a one-page agent estate report and understand what's there, what it costs, what it earns, and what's at risk.

If success is being measured in deployments, the role isn't working yet. The right unit is outcomes. "Customer complaint resolution time fell 30% and we retired 40% of our agents this year because the work changed" tells you something. "We have 200 agents in production" doesn't.

Invest early, or wait for the inflection?

My answer is yes, but staged.

Don't hire a full Chief Agent Officer on day one. There's plenty of work to do (the list above makes that obvious), but in most organisations the agent estate isn't yet at the scale or maturity where a standalone C-suite role can land properly. The executive team hasn't built the muscle of governing agents, the operating model is still forming, and dropping a new chief into that environment usually ends in a reorg or a quiet exit within the year.

Do start the work now. Find someone in your business who already understands both the agent estate and how the business actually runs, and make this their real remit. They'll sit under the CTO or COO today and won't need a new title yet. What they need is a proper scope: get a register of every agent in the business, work out what each one costs and what it's worth, give agents a real retirement process so taking one offline is a deliberate act, and have a route to brief the board when something agent-shaped goes wrong. The Citicorp lesson is that the time to build the muscle is before the first incident, not after.

Don't tie the promotion to a date. Look for one of these signals:

• Agent-augmented or agent-led processes touch revenue-critical workflows where a behavioural drift could be material.
• Total agent count crosses a meaningful fraction of headcount.
• The CFO starts asking unanswerable questions about the cost-to-serve of agent-heavy workflows.
• Your auditors, regulators, or insurers start asking who owns agent governance.

If you can answer "who owns this end-to-end" with a single name today, you're ahead of most of the leadership teams I tend to put the question to. If you can't, that's the work to start.

Why I think this matters more than another C-suite title

I'm aware of the cynical read. "Great, another chief, another fiefdom, another reorg." The C-suite has been growing for thirty years and not all of the additions have aged well.

What separates the roles that stuck (CISO, CDO) from the ones that didn't is whether the underlying domain became economically and operationally distinct enough that distributing it across the existing leaders genuinely stopped working. Information security got there. Data got there. I think the agent operating model is getting there now, faster than the leadership conversations I'm in suggest most companies have noticed.

Your business will eventually have someone in this role. The question is whether they arrive on your terms, before the first material incident, with the registry and the discipline already in place; or after, when you're recruiting at the same time as everyone else and trying to reconstruct what your agent estate has been doing for the last two years.

I know which side of that I'd rather be in the room for.