Sidestream reads the Milken Institute and Harris Poll survey published in Fortune on 20 May 2026 as a conversation gap rather than an AI gap: 68% of US workers feel they are navigating the AI transition on their own, and 41% received no employer AI support last year. The missing piece is the one-to-one in which a manager names what automation means for the person in the chair.
In Fortune on 20 May 2026, Karen Kornbluh (Senior Advisor, Milken Institute) and Libby Rodney (Chief Strategy Officer, The Harris Poll) released a survey reading that, taken in isolation, is one of the more uncomfortable workforce numbers of the year.
Sixty-eight percent of US workers say they feel they are navigating the AI transition largely on their own. Forty-one percent received zero employer AI support in the past year. Eighty percent want their government to start preparing workforce transition programmes now. The line between the people running organisations and the people doing the work has rarely looked further apart.
Stitched onto two other primary sources, the picture sharpens. The IMF estimated in January 2024 that about 60% of jobs in advanced economies may be impacted by AI.
A Stanford Digital Economy Lab study, revised in August 2026, finds that employment of workers aged 22 to 25 in AI-exposed occupations now stands 19% below where it would be had it kept pace with less-exposed peers. The transition is not theoretical. It is in the dashboards now, and the conversation about what it means for the person in the chair is largely not happening.
Why It Is A Conversation Gap, Not An AI Gap
The default reading of these numbers is to argue about the technology. Should companies roll out AI faster, slower, more widely. Should governments subsidise. The Milken-Harris finding is more uncomfortable: 88% of business leaders say no single company can solve AI workforce readiness alone, and 68% of leaders expect to be better off in five years, while only 27% of workers do.
The leaders know they cannot fix it on their own. The workers feel that. What is missing is the conversation about the part each manager, in each one-to-one, can still own.
That conversation has a behavioural shape. It is not a town hall, and it is not an FAQ. It is the moment a manager sits with a team member whose job has already been partly automated, names that fact, and works out together what the next twelve months looks like. Our coaching skills training in London rehearses conversations like this, with an actor playing the team member.
Gallup's 2026 State of the Global Workplace sits next to this directly. In US organisations investing in AI, employees whose managers actively support their team's use of AI are 8.7 times as likely to strongly agree that AI has transformed how work gets done. The manager layer is the lever. The Milken-Harris number is what happens when that lever is not pulled.
Edmondson's 1999 work, in Administrative Science Quarterly, defined psychological safety as the belief that speaking up will not lead to embarrassment, rejection or punishment. In a workplace where 41% can name no AI support and 68% feel alone, the calculus of speaking up about a deskilling job is exactly what Edmondson described.
Most people sit on it until they can no longer. The deskilling worry has data behind it: our post on AI skill atrophy and Microsoft's 2026 Work Trend Index shows the heaviest users switching AI off on purpose to protect their skills.
What Most Organisations Try
The default response is policy. An AI use policy, an AI ethics statement, an internal portal full of e-learning modules. The policy goes out. Adoption metrics are tracked. The conversation that would actually move the number, the one between a line manager and a team member about what their job is becoming, does not happen, or happens once badly and is not repeated. Deloitte's numbers on the AI trust crack point to a related gap.
This is a known behavioural pattern, not an HR failure. Roediger and Karpicke (2006), in Psychological Science, showed that being tested on material lifts long-term retention by roughly 50% compared with re-reading. A two-hour AI policy briefing is the re-reading kind. A manager rehearsing the actual sentences they will use with a worried team member, against feedback, is the testing kind. One of them holds under stress. The other does not.
Anders Ericsson's Peak (2016) goes further. The first-reaction behaviours that make a hard one-to-one land, naming the change in plain language, asking what the person is worried about before answering, sitting with the silence that follows, change only with deliberate practice. They do not change with insight.
What Actually Works
In the 2020 UCL study behind our method, rehearsal with a role-play coach was significantly more effective than passive modalities (Cohen's d = 0.75 to 1.08) at building communication skill, and self-rated confidence did not predict observed performance. We design the self-report out by measuring the behaviour. What that study does and does not prove is set out in the limits of our UCL study.
Applied to AI-transition conversations, the format is small group, one or two professional actors playing a deskilled administrator, a sceptical engineer, a quietly resentful long-tenure hire. Three or four full conversations, replayed with feedback against named behavioural anchors. By the third rehearsal the manager's first move is to ask the question, name the shift in the role, and stay in the silence. The 68% who feel alone do not feel alone because their manager is not trying. They feel alone because no one has rehearsed the conversation.
What We Do About It
Our immersive simulations and manager workshops are designed for exactly the rooms where the AI conversation has been postponed. Read also our piece on the Gallup 8.7x manager-AI multiplier for the manager-layer angle, or our Microsoft 26% piece for the leadership-alignment angle.
The 68% are not alone because the technology is moving fast. They are alone because the manager conversation has not been rehearsed, and policies are not conversations.
The organisations that hold their workforce through the next eighteen months will not be the ones with the cleverest AI policy. They will be the ones where every manager has rehearsed, out loud, the conversation a worried team member is already trying to start. Book a call to look at what that rehearsal would look like in your organisation.
Book a free 30-minute diagnostic call → or read about our research-backed approach.
The 68% Navigating AI Alone In Summary
The Milken Institute-Harris Poll names a workforce that knows the ground is moving and an employer that has not yet had the conversation. The gap is conversational, not technological. The lever is the manager who can sit with a team member whose role is changing, name it plainly, and work out the next twelve months, and that conversation only holds under stress once it has been rehearsed.
- 68% of US workers say they are navigating the AI transition largely on their own, and 41% received zero employer AI support in the past year.
- The IMF estimated in 2024 that about 60% of jobs in advanced economies may be impacted by AI, and a Stanford study revised in August 2026 puts employment of workers aged 22 to 25 in AI-exposed occupations 19% below the path of less-exposed peers.
- 88% of business leaders say no single company can solve AI workforce readiness alone, and 68% of leaders expect to be better off in five years against only 27% of workers.
- Gallup found that, in US organisations investing in AI, employees whose managers actively support AI use are 8.7 times as likely to strongly agree AI has transformed how work gets done.
Sources: Milken Institute and The Harris Poll, 2026 Listening Project · Gallup, State of the Global Workplace 2026