Executives Ownership in the Agentic Economy

The first working session had fifteen executives in it.

Their new CEO believes agents are becoming central to how buying decisions get made in his category, and he wanted to move. There's no AI officer, so when we asked who should be in the room, the answer was everyone with a claim to the question. Two VPs of Product, the CTO, the new VP of Data, the VP of Business Analytics, the CRO, the Chief of Staff, and the CMO.

Nobody owned "AI", so everybody came.

Look at that list again, because the composition matters more than the number. The chief revenue officer was there, and so was the CMO. Nobody convenes fifteen executives for an IT project.

Agents aren't buying. That isn't the point.

There's a comfortable dismissal going around: agentic commerce is overhyped, agents aren't completing purchases, so this can wait. The first half of that is true, and the conclusion still misreads how commerce works.

Cox Automotive ran the numbers in its annual car-buyer study last fall, the first year they've tracked it. A quarter of new-vehicle buyers - 25%, against 19% of buyers overall - used AI tools somewhere in their shopping.

Sit with that one. Americans spent about $620 billion on new vehicles last year, roughly 16 million of them, at an average sticker just over fifty thousand dollars. A quarter of the people spending that money had a model somewhere in the loop. Not buying through it - researching, comparing, narrowing, deciding. And almost nobody selling into that market can tell you which quarter, what the model said, or how to influence any of it. We spent twenty-five years learning to rank in a list of ten blue links. This is not that.

Transaction is the last step of commerce, not the whole of it. The money is won upstream, in the consideration set, and that's where agents live.

You also don't get to pick which buyer you serve. Both arrive at the same storefront - and if you sell through retailers or marketplaces you may not own the page the agent reads at all. One firm that tracks this found a brand's own site accounts for under 4% of what ChatGPT cites in its category. They don't behave alike either. I've spent months running controlled shopping trips with agents, and the coupons surprised me most: show an agent a discount and it didn't buy more, it shifted into comparison mode and read more competitors.

So who owns it?

Start with the easy case. When AI is a coding productivity question - ship faster, fewer defects - the owner is obviously the CTO. Someone owns the platform, someone owns the rollout, and you read the result by quarter's end.

Widen it a notch. When it's productivity across every function - sales writing differently, finance closing differently, legal reviewing differently - who owns that? BizOps has the process map. The Chief of Staff has the cross-functional mandate. The CTO still owns the tooling and the spend. One company I talked to put it under the CHRO, on the logic that it changes headcount and skills, so why not. All four are defensible, which is another way of saying none is obvious.

Go one notch further, to the thing that put fifteen people in a room. Preparing for an agentic economy isn't about your systems at all. It's about how your market behaves: how buyers find you, what they read, which claims they trust, what a machine makes of your product page. That lands on the CMO, the CPO, and the CRO at once, and none of them owns the customer's path end to end. B2B is the same shape, just slower.

Every prior technology wave had a natural home. Cloud went to IT, mobile went to product, data went to a CDO. Each was a system - something you could hand to a person and hold them accountable for. This isn't a system. It's a change in the behavior of your customers, and there is no chair labeled the market changed.

At the company I worked at before this, both halves were attempted at once, same building, same year. The market-facing half got a chief AI officer running a roadmap alongside the chief product officer. Two owners, one product surface. What followed was debate - thoughtful, well-argued, continuous - and very little shipped.

The internal half went to a tiger team out of business operations. And the moment a BizOps VP started booking time with functional leaders to interview them about processes that could be improved, everyone in the building heard the same word: headcount.

That's the awkward middle. Too big to ignore. Too small to rebuild around. Every transformation playbook ever written assumes total commitment, and nobody wrote the manual for twenty percent.

The answers today

The most popular answer to ownership is noone - decide this is somebody else's decade and wait.

When companies do move, they reach for what's available, and what's available is expensive:

Appoint a chief AI officer. The role is spreading fast - trackers of large US companies put adoption near 40%, a global survey of two thousand chief executives puts it at 76%. When estimates of whether a role even exists range that widely, the market hasn't agreed what it is. The postings confirm it. A typical one wants a production ML leader with eight to twelve years shipping models. Plus an enterprise strategist owning build-versus-buy and ROI. Plus a governance officer handling ethics, privacy, and regulatory exposure. Three jobs, one title, one salary line, commonly $750K to $1.5M - in the hardest category to hire globally. You're competing for a person who may not exist, to do a job nobody has specified, at a price that assumes they can.

Hire the consultants. Fine, and sometimes right, so long as you notice that AI advisory is now a multi-billion-dollar line at the largest firms, and an industry that size has a commercial interest in extending your dependence and drawing out the transformation.

Buy your way there. Every core system now ships AI features, and the pricing changed shape underneath you. Microsoft 365 Copilot runs $30 a seat on top of the base suite; Salesforce meters Agentforce by the conversation. These charge for what agents do, not how many people have logins, so you can't forecast the bill from headcount anymore. Your procurement team's whole muscle is price per seat, negotiate, multiply. It doesn't fit the shape of the thing.

Build the capability, not the bureaucracy

I've argued that as model capability commoditizes, durable advantage moves to the layer above it - the accumulated context and judgment nobody can buy off a shelf. That still holds, and this piece isn't the opposite of it. Building that capability and building a bureaucracy to own it are different acts, and only one compounds. A chief AI officer, a center of excellence, a multi-year contract are org moves: expensive, hard to reverse, and not the same thing as getting better.

So what does a capability move look like, if it isn't a hire?

Start with the question everyone asks first: how much of my business is agentic. Bots passed humans on the open web this June, but most of that is training crawlers with no buyer behind them - and the traffic is evaporating anyway, with two-thirds of US searches now ending without a click. "Did they visit" is becoming a question the web is built not to answer. Getting cited still matters, and an industry sells that now, but it's the visible half and it's crowded. The half nobody sells is what happened after: which claim the model repeated, which competitor it reached for, where it gave up. We know they're showing up. The open question is what they do.

That part is measurable. It's just hard to measure honestly, and I know because I've done it badly. In 3 test studies across internal marketplaces - three wrong answers, all of them plausible, all of them mine. The questions that help uncover whether a vendor/team is finding the truth to guide your strategy:

• What instructions did you give the agent, and what changes if one word of them changes?
• How did you pick what you measured, and what's the number across everything?
• How much of what you measured was your own software?

One each, for the three ways I've been wrong. And a fourth I'd want asked of me: can I reproduce this myself? You're the only auditor here with a real incentive to find the error.

None of it was caught by a vendor, a platform, or a consultant. It got caught because someone inside owned the question and kept asking. That's cheap, it compounds, and it survives being wrong - more than you can say for an org chart built on last year's number.

What I'm watching

I don't think the ownership question is solved, and I've come around to thinking it can't be yet - not the usual way, by drawing a box and putting a name in it.

The roles that stuck from earlier waves didn't arrive first. The CIO title spread after companies already had systems to run; the CDO showed up once there was data to govern. The work came first, and the role was named after it. We're running that in reverse, appointing owners for work whose shape nobody can describe. The closest thing to a right-sized structure I've seen is fractional, and I'd hold even that loosely.

One thing that cuts against me: I left to build in this space, so I'm betting the agent share compounds fast. I still think small and reversible is right, and the two aren't in tension. Speed is what makes an expensive, slow-to-unwind structure dangerous, because you don't yet know which capability you'll need. If the number doubles next year, instrumentation is how you find out before I do.

So the fifteen people in that room weren't a symptom of a broken company. They were the honest answer to a question the chart can't hold yet. The companies that come through this well won't be the ones that reorganized first. They'll be the ones that kept commitments small and reversible, got better each quarter at telling whether any of it worked, and drew the box only once they knew what belonged in it.