AI Stalls on Two Bottlenecks. Neither Is Technical.

On August 12, Claire Vo, CPTO of LaunchDarkly and the builder of ChatPRD, was asked for the top reasons AI stalls inside companies. Her answer started with the one almost nobody says out loud.

Read it twice, because it is not an insult. It is a job description that almost no company has filled.

The models work. That argument is over; the whole frontier now sits within a couple of index points of itself. The licences are bought: 85% of employees have access. What stalls is the step in between, and it stalls on two distinctly human bottlenecks. Neither is technical.

Bottleneck one: nobody whose job is reimagining the work

Using AI inside your existing workflow is easy, and it produces the 15% daily-usage number Gallup keeps measuring. Reimagining the workflow itself, deciding which steps stop existing, what the agent owns end to end, where the human moves from typing to deciding, is a different act entirely. It is product management performed on your own company.

And that is precisely the skill most companies never hired for internally. Product managers design workflows for customers. Nobody holds the same mandate pointed inward. So AI adoption gets assigned to whoever is enthusiastic, and enthusiasm without the reimagining skill produces faster versions of the same process, which is where pilots go to stall.

This is the pattern in the numbers we have covered before: the 15% who use AI daily redesigned their own work and nobody redesigned anyone else's. Garry Tan's version is that managers and CEOs have not greenlit radically different workflow plans, and until they do, macro gains take ten years rather than two.

Bottleneck two: the shadow org chart

Suppose you have the rare person who can reimagine the work. The redesign still has to be adopted, and adoption is decided by whoever the CEO actually listens to, which is not the same thing as the org chart.

The same day, coincidentally, and the same subject from the other side. A workflow redesign that never reaches the shadow org chart dies in a deck. One that does gets greenlit in a week. Most AI initiatives are staffed for neither bottleneck: the team can build, but it cannot reimagine, and it reports to someone who cannot change the CEO's mind.

What clearing both actually takes

Treat workflow reimagination as a role, not a hope. Somebody must hold the explicit mandate to redesign how a function works, with the product skill to do it. If nobody inside the building has both, that is a hiring gap or an embedding gap, not a tooling gap.

Place the work inside the shadow org chart, not beside it. Before an AI initiative starts, name the person who can change the CEO's mind, and make them the sponsor who reviews the redesign, not a recipient of its summary. If you cannot name that person, the initiative is already dead and merely has not noticed.

Ship the redesign one function at a time. The reimagining skill compounds: the second function is faster than the first, because the pattern of deciding what stops existing transfers even when the workflows do not.

The reason AI stalls is not in the model card. It is a missing role and a missing sponsor.

Staff both.