The Deck Nobody Wrote

The Deck Nobody Wrote

Status
Draft
Author
Created
Jul 7, 2026
Tags
Agentic AI
Consulting Ethics
Accountability
Text
When an agent builds the model overnight, accountability doesn't vanish. It just moves.
A senior associate walks into a client review at eight in the morning. She slept seven hours. The fifty-slide market-entry model on the screen was assembled overnight by an agent, working from three data feeds and a prompt she wrote before dinner. She stands up and defends every number on every slide. The client pushes back on a growth assumption in slide twelve; she reworks the logic live, explains the tradeoff, and moves on. Nobody in the room asks who built the deck. They ask if it's right.
This scene is no longer rare. McKinsey's most recent read on enterprise AI adoption found that 62% of organizations are now experimenting with agentic AI, and 23% have begun scaling agents into at least one business function. The research keeps circling back to a single unresolved question: who is accountable when an agent's output is wrong? The current answer, in most large organizations, is procedural — assign an owner, log the decision, write it into a responsible-AI framework, cite ISO 42001. That's a governance answer. It isn't the answer the client is actually asking for when they push back on slide twelve.
Here's the principle underneath the scene. For most of the history of professional services, authorship and accountability were the same thing, because they had to be. If you built the model, you were the only person who could defend it — which meant the act of building it forced you to understand it. Rigor wasn't a virtue consultants chose to have; it was a side effect of the fact that nobody else could stand in for you when the client asked why. The hours you billed were, in a rough way, a proxy for the thinking you'd done.
Agents sever that link cleanly. You can now defend excellent work you didn't build. You can also defend mediocre work with exactly the same confidence, because the fluency of the presentation no longer correlates with the depth of the thinking behind it. A deck that took an agent four hours to assemble looks identical, at first glance, to one a senior partner labored over for a week. The tell used to be in the seams — the slightly-off formatting, the inconsistent voice, the places where a junior analyst's inexperience showed. Agents don't leave those seams. Which means the old proxy for trustworthiness — how hard did this look to make — has quietly stopped working, right as the volume of AI-assisted output is exploding.
Some firms are reaching for a replacement metric already. The Human-Agent Ratio — how many agents a given employee is effectively directing — is gaining traction as a way to quantify scale. It's a useful operational number. It is a bad proxy for accountability, for the same reason headcount was always a bad proxy for judgment: it counts capacity, not discernment. An analyst directing twelve agents who rubber-stamps their output owns nothing more than one directing a single spreadsheet macro.
What actually transfers responsibility from agent to human was never authorship. It was always judgment — specifically, three judgments the agent cannot make for you. What question got asked, and what got left out of the prompt. Which of the agent's outputs got trusted and which got quietly discarded, and why. And whether you're willing to stand in a room and be wrong about it in front of a client who's paying for your name, not your keystrokes. That third one is the whole job now. Field experiments on human-AI teamwork are starting to confirm what this scene already suggests: the teams that outperform aren't the ones that generate the most output, they're the ones whose humans exercise the most discerning judgment over what the agent hands them.
Which means the billing model built around hours and deliverables is pricing the wrong thing, and has been for a while now — agents just made the mismatch impossible to ignore. A client paying for a hundred hours of "analysis" when an agent did the analysis in four is paying for theater. What they should be paying for is the fifteen minutes where a person decided which three scenarios out of forty were worth the client's time, and the further ten minutes where that same person decided to walk into the room and defend a number they'd bet their reputation on. Consulting engagements that survive the next decade will be the ones that figure out how to price that — not the hours, the nerve.
The associate in that opening scene didn't write the deck. But she signed it, in the only way that has ever mattered — by being willing to be wrong about it, out loud, in front of the person paying for the answer. That's not a lower bar than authorship used to be. It might be a higher one.