TalentVector NextPass ↗

Evidence

It fails at the manager

September 2026 replicated

Adaptation does not spread on its own. One person changes how they work, and either that becomes how a team works or it stops with them. Management is the layer where that transfer happens or doesn’t — where an individual habit becomes an expectation, then a norm, then how the work gets done here. Every organization has that layer. Almost none observe it.

Four independent 2026 studies, on different samples with different methods, land on the same shape: managers are asked to carry that transfer and have not been equipped to. The pattern replicates. Below is what each one measured — and, at the end, what none of them establish.

Managers are already further ahead. That is the problem.

Gartner, March 2026: 46% of managers say they are experimenting with AI to improve their work, against 26% of individual contributors — roughly 1.8 times the rate (n=2,986 employees). Managers are both out ahead of their teams and accountable for closing the gap behind them. Nothing in that survey says they were given anything to close it with.

A second Gartner survey in the same March 2026 release sharpens the point. Just 7% of organizations provide any guidance on what to do with the time AI frees up — n=114 HR leaders, a small sample, and we label it that way. Hours get returned to the workday and nobody says what they are for. Hold that figure while reading the rest: the gap starts upstream of the manager. Nobody handed them a standard.

Expectation without accountability

Section’s AI Proficiency Report (June 2026, n=5,026 US knowledge workers) found that 65% of managers either set no expectation for AI use, or set one with no accountability behind it.

The compound matters. A meaningful share of that 65% did encourage AI use, and then nothing followed it: no standard, no follow-up, no difference either way. An unenforced expectation is a failure mode, not indifference — it is what happens when someone is asked to endorse a change they have no way to see. You cannot hold anyone accountable to a behavior you cannot observe.

Confidence, measured directly

The Chartered Management Institute — a professional body rather than a vendor — surveyed more than 1,000 UK managers in June 2026. 12% said they feel “very confident” managing teams that use AI. Asked about agentic systems, 10%.

Note what kind of evidence that is: a confidence survey — rung three on the Evidence Ladder, where someone says they feel prepared. It is the most independently sourced figure here, and still an answer to a question rather than an observation of a behavior.

The gap is invisible from above

Acorn’s 2026 State of Learning for AI Fluency (May 2026, more than 1,200 professionals) reports 77% of executives believing their managers are prepared to guide AI skills development, against 91% of employees who say they are not.

Two caveats. The claim is narrow — guiding AI skills development, not leading AI-era work generally — and should not be widened. And its provenance is the least firmly established of the four here: the figures come from a syndicated release rather than the report itself, and the sample detail was not independently verified.

At its own scope it says something the other three cannot: the layer above the manager does not know. Executives are describing a manager population that employees do not recognize.

What this does not show

It does not show that managers are the binding constraint on AI-era work.

No source above ranks managers against the other candidates — tool access, data quality, incentive design, workflow redesign. None of the four discusses any other constraint at all: they are HR and L&D research, and the manager is the subject they were built to examine. Stacking them produces convergence about managers largely because managers were the only thing measured.

Where researchers have looked comparatively, the emphasis falls elsewhere. McKinsey, BCG and Deloitte all point at redesigning the flow of work. Deloitte’s State of AI in the Enterprise 2026 (n=3,235 leaders across 24 countries) reports that 84% of organizations have not redesigned jobs or workflows around AI. That figure is verified, and it is evidence for a competing explanation rather than ours. It belongs here for that reason.

The narrower argument that holds

Of the candidate constraints, the manager layer is the least observed. Workflow redesign has consultancies, process maps, and org charts pointed at it. Tool access is a license count. Manager behavior — what a person expects, models, and follows up on with the individual in front of them — is established, in every study above, by asking someone about it afterward. Every figure in this note is a survey answer.

That is the argument. Not that this constraint is the largest, but that it is the one we have the least direct evidence about — and that the shortfall is a measurement problem, not a fact of organizational life. Behavior between a manager and their team is spoken out loud. Spoken behavior can be observed against a standard chosen in advance. Observability is the part that can be fixed.

An organization can buy every license and redesign every workflow and still find the change stops one level above the work — because between a person’s adaptation and their team’s, there is a conversation somebody has to have. Management is that conversation. Whether it is the largest constraint is unsettled. That it is the one nobody is watching is not.

Sources

← All evidence notes