Profound's End-to-End Marketing Agent and McKinsey's Healthcare Crisis Expose the Same Delegation Bias

Two stories broke in the same week that, on the surface, share nothing. McKinsey published a report on healthcare's productivity crisis, arguing that more hiring and more technology won't fix a system stretched to its limit. The fix requires redesigning human-AI workflows from scratch. Profound launched an AI agent that promises to manage marketing end-to-end, from strategy through execution, while quietly flagging that rising token costs could complicate the rollout.

A hospital system and a martech vendor. Different industries, different customers, different vocabulary. But read them back to back and you notice they're both describing the identical unresolved question: which decisions should a human make, and which should a machine execute? Neither company has actually answered it. McKinsey admits the industry hasn't figured out the workflow redesign. Profound is selling the answer before it exists.

That gap (between claiming a decomposition problem is solved and actually solving it) is where I want to spend this essay, because I think it's the central strategic risk of 2026, and almost nobody is naming it correctly.

Two Industries, One Unsolved Decomposition Problem

Healthcare's productivity crisis isn't a staffing shortage in the conventional sense. McKinsey's point is sharper than "we need more nurses." It's that the current system asks humans to do work that automation could absorb, while automation gets deployed in places that still require human judgment, and the mismatch is what's actually driving burnout and cost. The fix isn't adding headcount or adding software. It's correctly assigning each task in the workflow to the party (human or machine) best suited to do it.

Marketing has the exact same structural problem, just dressed in different language. Profound's pitch is that an agent can run marketing "end-to-end", meaning it doesn't just execute campaigns, it makes strategic calls about what to run and why. That's a bigger claim than automating a workflow. That's automating judgment. And Branding Strategy Insider's recent piece makes the counter-case plainly: AI can generate ideas, optimize campaigns, and personalize experiences, but it cannot decide what a brand should stand for, which opportunities are worth pursuing, or which promises should stay non-negotiable. Those are judgment calls, not optimization problems.

So you have two live examples of the same unresolved question: McKinsey saying "we haven't cracked the decomposition yet, and it's costing healthcare dearly," and Profound saying "we've cracked the decomposition, buy our agent," while the branding trade press is quietly arguing the decomposition Profound claims to have solved is the one thing that can't be automated at all.

I've written before about how McKinsey's own agentic AI research contains a structural failure diagnosis that most marketing orgs ignore, and how IKEA's prioritization problem with agentic AI traces back to the same root cause: nobody built a principled rule for what gets delegated before the tooling showed up. What's different this time is that I think I can now name why that rule keeps not getting built. It's not a workflow design gap. It's a psychology gap.

This distinction matters because the standard enterprise response to the decomposition problem is structural: map the workflow, identify the handoffs, install the guardrails. That approach fails not because the mapping is wrong, but because the guardrails get designed by the same people who have motivated reasons to draw the line in the wrong place. You cannot audit your own Homobiasos. The executive who is most threatened by agentic AI is the least qualified to define what it should and should not control, and they are usually the person making that definition.

The Reasoning Isn't Broken. It's Protecting Something Else.

BehavioralEconomics.com's recent piece on what it calls "Homobiasos" makes an observation that should unsettle anyone who thinks their org's AI adoption decisions are being made rationally: human reasoning often doesn't exist to find the truth. It exists to protect the reasoner's sense of self. We don't rationalize because we're bad at logic. We rationalize because accurate conclusions sometimes threaten our identity, our authority, or our emotional stability, and the brain will happily sacrifice accuracy to preserve those things.

Apply that lens to how executives are actually deciding what to delegate to agentic AI right now, and the picture gets uncomfortable fast. Two failure modes show up constantly, and both get dressed up as strategic reasoning when they're really self-protection wearing a lab coat.

Failure mode one: over-delegation as avoidance. Handing an agent a decision not because the agent is qualified to make it, but because making it yourself is uncomfortable. It requires taking a position, defending a tradeoff, owning a call that might be wrong. Calling this "efficiency" or "scaling the team" is the rationalization. The actual driver is that hard decisions are unpleasant, and an agent never pushes back in a meeting.

Failure mode two: under-delegation as ego protection. Refusing to hand off genuinely mechanical, low-judgment execution work (scheduling, reporting, basic optimization) because relinquishing that control threatens the leader's sense of being "the strategist" or "the person who knows the account." Calling this "brand stewardship" or "quality control" is the rationalization. The actual driver is that giving up visible busywork feels like giving up relevance.

Neither failure mode is really about the AI. Both are about identity protection, and Homobiasos's core claim is that this kind of self-protective reasoning happens beneath conscious awareness. Leaders genuinely believe the story they're telling themselves about why they delegated what they delegated. That's what makes it dangerous. It doesn't feel like bias from the inside. It feels like judgment.

The Delegation Rationalization Trap

Here's the framework I'd propose, and it's the piece none of these five source articles state directly: organizations aren't failing to adopt agentic AI because of cost, complexity, or workflow design. They're failing because the line between "delegate this" and "don't delegate this" gets drawn by whichever bias is loudest in the room that week: avoidance one quarter, ego protection the next. Both get retroactively justified as strategy.

You can see this pattern show up in the healthcare data McKinsey cites and in the marketing agent Profound just launched, from opposite directions. Healthcare systems have been slow to delegate mechanical documentation and triage work to automation. Not because automation can't do it, but because clinical culture ties professional identity to doing everything by hand, and delegating feels like devaluation. That's under-delegation as ego protection, at industry scale. Marketing orgs adopting tools like Profound's agent risk the opposite error: delegating brand-defining judgment calls (what the brand stands for, which markets matter, which tradeoffs are non-negotiable) to an agent, because making those calls is genuinely hard and an agent will make one instantly. That's over-delegation as avoidance, and it's exactly the terrain Branding Strategy Insider is warning about.

This connects to something I've argued before: the real work of brand strategy is coordination, not communication: deciding who does what, in what order, and why. Agentic AI doesn't remove that coordination problem. It raises the stakes of getting the decomposition wrong, because now the wrong assignment gets executed at machine speed and machine scale.

The Pre-Commitment Test

If motivated reasoning is the mechanism, the fix isn't "think harder about it in the moment." Homobiasos's whole point is that in-the-moment reasoning is exactly where the bias lives. The fix has to happen before the emotional stakes are attached to a specific decision.

Here's a test I'd propose: before deploying any agent (marketing, clinical, operational), write down, in advance, the criterion for what counts as "judgment that stays human" versus "execution that gets delegated." Not case by case. As a rule, stated before you look at any specific outcome. If you can't articulate that criterion in writing, in advance, and defend it to someone who has no stake in the deployment, you're not making a decomposition decision. You're rationalizing one after the fact based on how each option makes you feel.

The mechanism here maps directly to what psychologists call "motivated reasoning" at the organizational level. Individual leaders may be aware of their biases in the abstract, but awareness doesn't resolve the problem. The executive who knows about Homobiasos is still operating inside an identity structure that the AI deployment is threatening. Structural solutions have to do what individual awareness cannot: create formal, documented, time-stamped pre-commitments that exist independently of the emotional state during deployment. A decision log that records the delegation rule before the vendor is selected is harder to rationalize away retroactively than a rule that lives only in someone's head.

This is the same discipline I wrote about in Stop Delegating the Core: the danger isn't automation itself, it's automating the thing that defines you without noticing you've done it, because noticing requires exactly the kind of uncomfortable self-scrutiny Homobiasos says the brain is built to avoid.

What This Costs, Specifically

Profound flagged rising token costs as a complication for its agentic rollout, and I think that detail is more important than it looks. Vendors tend to frame token cost as a scaling problem: more tasks, more tokens, linear math. I'd argue the real cost driver is misdelegation, not volume. When an agent is handed a decision it's structurally unqualified to make (a brand-defining tradeoff dressed up as a campaign optimization), the output triggers revision loops. Humans reject it, re-prompt it, re-scope it, re-run it. That's where token costs actually compound: not in the execution work agents are good at, but in the judgment work they're bad at and keep getting handed anyway.

If that's right, the token cost line item on next year's marketing budget isn't really an infrastructure expense. It's a receipt for every time an organization's leadership avoided drawing the delegation line honestly and let the bias draw it instead.

Even outside the enterprise, this pattern isn't unique to boardrooms. Look at something as small as Alaska Airlines' Atmos card update: two nearly identical cards with different terms, and consumers routinely pick based on identity ("I'm an Alaska flyer") rather than running the actual math on which offer serves them better. Same rationalization mechanism, much lower stakes. The instinct to let identity substitute for analysis doesn't discriminate between a $95 annual fee and a nine-figure automation strategy.

The uncomfortable conclusion is that the skill brand leaders actually need in 2026 isn't AI literacy or prompt engineering. It is the discipline to write down, before deployment and before the emotional stakes attach, a falsifiable rule for what agentic systems are and aren't allowed to decide. Precisely because your reasoning about that boundary, in the moment, cannot be trusted to be about anything other than protecting how the decision makes you feel.

If your organization is deploying agentic AI into marketing, operations, or customer-facing workflows and hasn't written that rule down yet, that's the actual starting point, not the vendor selection. We've built research specifically on where that decomposition line tends to get drawn wrong; you can see it at smarttechinvest.com/research.