CEOs Own the AI Budget. Who Owns the Agent's Actions?

A LinkedIn post attributed to a consultant named Paul Shirer, titled "The State of the CEO on AI: Q3 2026," is what put a specific pair of numbers in front of me: 72% of CEOs say they own the AI decision, and, alongside it, a claim that only 14% can name the P&L line that decision moves. I could not reach linkedin.com from the environment I researched this in, and repeated searches could not independently confirm the post's text, its author's exact claims, or the post's continued existence beyond a search engine's own summary of it. Nothing from that post appears as evidence anywhere below. The 14% figure in particular traces to nowhere I could verify: no named survey, no methodology, no denominator. I am naming it here only because it is what sent me looking. It does not appear again in this record.

The 72% figure does trace to something real. It is the headline finding of BCG AI Radar 2026: As AI Investments Surge, CEOs Take the Lead, the third annual edition of Boston Consulting Group's global AI survey, published 15 January 2026. BCG surveyed 2,360 executives across 16 markets and nine industries, including 640 CEOs specifically. "Nearly three-quarters of chief executives (72%) say that they are now the main decision makers on AI," BCG's press release states, a figure it describes as "twice the share as last year." That is a self-reported figure: CEOs describing their own role, not a measurement anyone else made of them.

What the primary evidence establishes

Set aside the LinkedIn post entirely. Here is what BCG's own published material for AI Radar 2026 actually supports, read for exactly what it measures and no further.

CEOs increasingly describe themselves as the main decision maker on AI, at 72%, roughly double the share BCG reported a year earlier. Half of CEOs say they believe their own job depends on getting AI right. Ninety four percent say they will keep investing in AI at current or higher levels even if it does not pay off in the next year. Companies are planning to roughly double AI spending in 2026, to about 1.7% of revenue on average, more than twice the increase BCG recorded for 2025, with 2026 spending plans ranging from 0.8% of revenue at industrial and real estate firms to 2.1% at tech companies and 2.0% at financial institutions.

Inside that spending, agents specifically are taking a growing share. CEOs have committed more than 30% of their organizations' 2026 AI investment to agentic AI. BCG groups CEOs into three self-reported archetypes: Followers, about 15%, cautious and early stage; Pragmatists, about 70%, investing only where value and risk are already clear; and Trailblazers, about 15%, investing decisively and upskilling fastest. Trailblazers direct more than half of their 2026 AI investment to agents, and are about twice as likely as Followers to deploy an agent end to end across a whole workstream or process rather than one step of it. Ninety percent of CEOs say they believe agents specifically will produce measurable ROI in 2026, a belief BCG records, not a result it measured. Confidence that AI will pay off varies widely by region too, from 44% in the UK to roughly three quarters of CEOs in India and Greater China, which is itself a sign that "CEOs believe AI will pay off" is sentiment the survey captured, not a financial outcome anyone has confirmed.

What it does not establish

Nowhere in BCG's published material for AI Radar 2026 is a CEO asked who authorizes an individual agent's action, what resource or system that agent is permitted to touch, how long that permission lasts, or whose sign-off a specific action requires. "Main decision maker on AI" is a claim about strategic and budget ownership: who sets direction, who approves the spend. It is not a claim about operational control at the point an agent actually acts. The same report that finds CEOs committing over 30% of a growing AI budget to agents, and its most aggressive adopters pushing that past half while running agents end to end across a workstream, does not ask whether anyone in that organization can say what a specific agent was allowed to do, or who answers for it if the agent exceeds that.

I want to be precise about what this absence means and does not mean. BCG's report is not making a claim about execution authority, governance, or accountability one way or the other. It simply was not built to measure any of those things. That absence is still informative on its own terms: the same organization tracking how confidently CEOs claim ownership of the AI decision has, in its newest published research, nothing to say about whether that ownership extends down to the individual actions being taken under it.

Why it matters

Put the two verified facts next to each other rather than reading them separately. CEO-claimed ownership of the AI decision has roughly doubled year over year, to 72%. In the same report, the share of AI investment CEOs are directing specifically at agents, systems that act rather than only generate, has passed 30% overall and half among the CEOs pursuing this most aggressively, the ones BCG calls Trailblazers, who are also about twice as likely to let an agent run end to end across a process rather than stopping inside one gated step of it. Claimed ownership of the decision and the volume of authority being handed to agents are rising together, in the same report, from the same executives. Nothing in that report addresses whether the mechanism for granting an individual agent's authority, scoping it, and later reconstructing what it was actually allowed to do is rising anywhere near the same pace.

Connection to execution authority

This is the layer Moona Intelligence keeps returning to. A decision made once, at the top, does not automatically answer who authorized one specific action, taken by one specific agent, against one specific resource, at one specific time. Approving an AI budget is a decision about direction and spend. Whether an agent's authority actually gets checked at the moment it acts is a different decision entirely, made differently, usually by different people, and often not made explicitly at all. BCG's own numbers show why that second decision is getting harder to leave unmade: more budget, a growing share of it aimed at agents specifically, and among the most aggressive adopters, agents already running end to end rather than one step at a time with a person in the loop between steps.

A CEO who can accurately say "I approved this budget" has answered a real question, and BCG's survey shows most CEOs now believe they can answer it. It is a separate question, unaddressed by anything in this survey, whether that CEO or anyone else in the organization can accurately say who approved the agent's last action, what it was permitted to affect, and whether it stayed inside that boundary. Rising investment in agents that BCG's own Trailblazers now run end to end raises the cost of leaving that second question unanswered. It does not, on this evidence, mean anyone has started answering it.

What to watch

A few specific things would tell me whether this gap is closing, none of which BCG's current survey addresses:

Whether BCG's next AI Radar, or a comparably resourced survey, asks CEOs a question about agent-level authorization rather than budget ownership or investment sentiment, and what it finds if it does. Whether organizations already reporting agentic investment above 30% of AI spend, which per these figures is already the described norm, say anything publicly about how they assign or record who authorized a given agent's action. Whether boards begin asking for agent-action-level accountability as a distinct line item from AI budget reporting, given that half of CEOs in this survey already say their own job depends on getting AI right overall. Whether a future edition of this kind of survey separates deployed, acting agents from AI adoption in general, since this edition's questions are almost entirely about investment plans and belief rather than what is actually running and what it is allowed to touch.