AI Proficiency Isn't the Bottleneck. AI Judgment Is.
Deciding which tool to use is much less important than what you’ll actually deliver with that AI tool
Section AI just released a report called The AI Proficiency Report. It's worth reading. The headline number — that 85% of enterprise AI use cases generate no business value — is real, and the company deserves credit for putting hard data behind something everyone in product management has been sensing for a year.
The number is right. The diagnosis isn't.
The report reads that 85% failure rate as a workforce-literacy problem. Only 9% of workers are "AI-proficient," it says. 75% feel overwhelmed. 43% are using AI tools against company policy. The implied fix is more training, more certifications, more proficiency. Lift the 9%, the argument goes, and the 85% waste drops.
That's not what we see in the work.
The pattern is older than the tooling
Josh and I have spent the last fifteen-plus years inside product organizations in healthcare, payments, retail, government, and education. Josh wrote Outcomes Over Output in 2017 because we kept watching the same thing happen, project after project, company after company. Teams would ship things — beautifully built, technically impressive, fully released — that nobody used. Not because the teams couldn't code. Not because they couldn't follow the framework. Because no one on the team had a defensible answer to a simple question: what customer behavior is going to change because we shipped this?
That gap has nothing to do with tool literacy. It's a discipline gap.
AI hasn't changed that pattern. It's made it faster and more expensive to repeat.
You can be the most AI-proficient PM in your company. You can tokenmaxx every prompt, run every framework on every roadmap, ship every feature your model spits out. If your customers don't change their behavior because of what you built, the work was free. AI made the bad outputs faster — that's all. Training the workforce on better prompts will accelerate the same problem.
What's actually scarce
Learning the tools is the easy part. A few weeks. Anyone motivated can do it.
Learning what's worth shipping is the hard part. It's not a skill you certify. It's experience plus customer fluency plus a feedback loop that punishes wishful thinking. It's the muscle that lets a PM look at an AI feature in development and say "this won't change anything customers care about" before the team ships it, and back the call up with something more than gut. Some people now call this taste. Whatever the label, it's the actual scarce skill the 85% failure rate is exposing.
Three questions every product team should be able to answer about every AI feature on their current roadmap:
What customer behavior needs to change for our work to matter?
Which feature actually serves that change?
Which feature is theatre?
Most product teams cannot answer those reliably. That's the gap. It's not closed by another certification.
It's closed by repetition. Talking to customers. Watching what they actually do. Killing work that doesn't move them. Defending work that does, even when it's unpopular. Doing it again next sprint.
Where Sense & Respond Learning fits
Sense & Respond Learning is the company Josh and I founded to teach exactly that discipline. We run Lean AI workshops for product teams trying to get judgment back into AI roadmaps. We run Outcome-Centered AI sessions for leadership groups trying to align an org around the questions above. We run a Certified Trainer Program for in-house practitioners who need to do this work at scale inside their own companies. Our work shows up across healthcare, payments, retail, government, and education — every domain where shipping the wrong AI feature has a real cost.
If you are the PM staring at an AI roadmap you don't trust, more proficiency training is not the move. The move is to walk the team through one feature, slowly, and force the team to answer the three questions above. If they can answer them, ship it. If they can't, find out what evidence would let them — and go get it.
That's the work. It hasn't changed in fifteen years. AI just made the cost of skipping it more visible.