A consulting partner I know presented a 48-slide strategy deck last month to a Fortune 100 board. Her AI agent had assembled the bulk of it overnight — competitive analysis, financial modeling, executive summary, speaker notes. She reviewed it, sharpened two sections, added one slide, and walked into the room.
The board approved the strategy.
Was it her work?
Her instinct was yes. She directed the research. She reviewed every claim. She made the call to present it. She would catch the consequences if it turned out to be wrong. In every functional sense, the deck was hers — she owned the outcome.
But somewhere beneath that instinct sits an uncomfortable question. If the agent had written it slightly differently, she would have presented something slightly different. If it had missed a risk, she likely would have too — not because she's careless, but because the review process is only as good as what you know to look for. The agent's fingerprints were everywhere. Her fingerprints were mostly in the approval.
This is the accountability problem that enterprise AI governance is only beginning to reckon with. A recent Altimetrik study found a stark "AI Transformation Gap" — widespread agent deployment across enterprises, with dramatically unclear ownership of what those agents produce. Organizations are great at delegating to agents. They're poor at being honest about what that delegation actually means.
Execution vs. Judgment
The instinct to say "yes, this is my work" isn't wrong — it's just incomplete. It rests on a definition of ownership that consulting has always used: ownership means accountability for outcomes, not credit for execution. Clients don't pay for hours of someone typing. They pay for judgment — about what to look for, how to frame a problem, when to challenge an assumption, what risks actually matter.
By that definition, the partner owned the deck. Her judgment shaped what the agent was asked to find. Her experience told her which two sections needed sharpening. Her read of the room determined what to leave in and what to cut.
But this is where it gets uncomfortable: judgment doesn't arrive fully formed. It's built through execution. Through the nights you spent writing a bad first draft and seeing where your thinking fell apart. Through the models you built wrong before you understood why they were wrong. Through the analyses you ran yourself and found the thing that didn't quite add up.
The McKinsey associate who drafts a hundred decks under a senior partner isn't just a pair of hands — they're building intuition. Learning to notice when a slide obscures instead of clarifies. When a narrative has a hole in it. When the data is technically accurate but emotionally misleading. That intuition is the output of execution, not a substitute for it.
The Apprenticeship Trap
The emerging operating model for agentic AI — what some engineering teams are calling "delegate, review, and own" — is structurally sound. Agents handle first-pass execution: research, scaffolding, drafting, testing. Humans review, make decisions, own the outcomes. Clear accountability. Scalable autonomy.
What this model doesn't solve is the pipeline problem. If junior consultants, analysts, and engineers spend their formative years reviewing agent output rather than producing their own, what does their judgment run on? Review skills are different from production skills. Catching errors in someone else's work — or something else's work — is not the same as knowing how to avoid making them.
Microsoft's 2026 Work Trend Index suggests the workforce is reshaping into an hourglass: strong demand at the senior level (people who can direct and judge) and at the junior level (people who support the infrastructure), with the middle — the traditional apprenticeship tier — hollowing out. That might be economically efficient. It is almost certainly a judgment-production problem in slow motion.
What Ownership Actually Requires
So when the partner asks "is this my work?" — the honest answer is: it depends on what work you've been doing before today.
If she built her judgment through years of first-principles analysis, the agent is a force multiplier. Her ownership is real. Her review carries weight because she knows what she's looking for.
If she has spent the last five years reviewing agent output without producing her own, her review is thinner than it feels. The approval is hers. Whether the understanding is hers is a different question.
The governance frameworks being built right now — the ADLC, the accountability matrices, the "who gets paged when the agent fails" org charts — answer a necessary but insufficient question. They tell you who's responsible. They don't tell you whether the person responsible has the depth to catch what an agent gets subtly wrong.
Ownership isn't a role you assign. It's a capacity you build. And we are designing workflows that may be quietly eroding the very process that builds it.