Icky vs Tricky: A hidden barrier to agent adoption

Your agent has a great business case. Strong ROI. And it works.
But nobody's using it. Harvard just mapped why.

23,570 people, 940 occupations.

The finding: we're not afraid of AI taking jobs—we're afraid of AI taking the wrong jobs.
Americans support automating 30% of occupations right now.
Tell them the agent outperforms humans at lower cost? That jumps to 58%.

But 12% of jobs are off-limits.
Caregiving.
Therapy.
Spiritual leadership.
Almost no one wants this.

88% of resistance is Tricky—performance problems.
"Does this actually work yet?"

12% of resistance is Icky—principle problems.
"Should this even exist?"

McDonald's scrapped its drive-through AI after viral videos of customers screaming "STOP" as the system racked up 260 McNuggets.
That's tricky.
Fix the tech, try again.
(They are.)

Therapy chatbots?
The American Psychological Association met with the Federal Trade Commission to regulate them after multiple wrongful death lawsuits.
That's icky.

Experts say "AI cannot replicate genuine human empathy" and no amount of capability improvement changes that objection.
Companies burn money when they treat Icky like Tricky.
Thinking better models will crack it.
Like pushing harder on a door that says PULL.

You can't engineer your way out of a moral boundary.

—THE TEST—
Imagine your agent is 10x better than any human at this task.
Does resistance vanish?

If YES: The problem is Tricky.
Keep at it. Tech’s always getting better.

If NO: The problem is Icky.
The culture's not there yet. Avoid.

—WHERE TO START—
The agent economy isn't constrained by what's technically possible or economically viable. It's constrained by what people will actually accept.

Healthcare shows this: AI clinical documentation hit 100% adoption and 53% report high success. AI mental health support? Only 39% interest among older adults.
Same industry, opposite results.
One's tricky, one's icky.

Winners aren't solving the hardest problems.
They're solving problems where performance improvements unlock adoption.
Where getting better is enough.
That's where pilots become platforms instead of PowerPoints.
The math doesn't matter if no one uses it.

Where are you seeing this in your agent deployments?