What the Room Knew

We spent last month talking to 25 young people (18-30) across ten
countries about AI and the future of work. We wrote it up and
published. Then we held our first ever Full Moon webinar - and
unsurprisingly and happily got deeper and further with a subject that
concerns many of us.

The three participants who joined us on the panel - Fernanda, Fergus
and Skyla - said things live that they had also said in the research
sessions. But said them differently. In front of 75 people, with two expert
practitioners Sophie Wade and Amanda Schneider, and 90 questions in
the queue, the ideas sharpened. New things emerged. Some of what we
had written was confirmed in ways we hadn't expected. Much of it was
insightful additional nuance.

This is our record of what the room added. Thank you to everyone who
turned up and especially to our excellent panel.

Please do share with anyone who couldn't make it - all of our content is now free because we want this thinking to reach as many people as possible.

What we already knew - confirmed from the inside

Ambivalence was the dominant finding in the written research. We had expected to find either optimism or pessimism. Instead we found both at once, often in the same sentence and from the same person.

That ambivalence perhaps surprised even the participants themselves. Fergus said it directly:

The only thing that surprised me is that it was more ambivalent rather than negative. I always assumed from my peers that it is a more negative space - and seeing that it was more 50/50 was fairly surprising to me - Fergus, product manager, London.

That gap between the private individual experience and the assumed collective mood is itself a finding. We may be suffering from a kind of secondhand anxiety. The story we tell ourselves about how this generation feels may be darker than how they actually feel.

The internal dialogue - a finding we missed

One thing the webinar surfaced that isn't in the written research: the negotiation people are conducting with themselves.

Skyla described stopping using AI for idea generation. Choosing to stop - because she noticed that she was reaching for a LLM before she'd really driven out her own thinking, and that this was hollowing out the work:

I started finishing tasks and then thinking: what was actually my idea in this? I finished it, but what have I actually learned? - Skyla, student, University of Edinburgh.

She is now in her final year. Her entire degree grade rests on the year ahead. Some of her peers will use AI more aggressively. She's going to hold the line anyway - 10 minutes, pen and paper with her own thinking first.

We wish this had been a named finding in the essay! It seemed to speak to a lot of participants - and us. What Skyla is describing active, deliberate self-governance - conducted alone, against real competitive pressure, without institutional support. It is a form of agency that the research pointed toward but never quite named.

It is also, we suspect, far more common than the people doing it realise. 

Frictionlessness is not always good design

The credibility gap - the worry that junior people are building deliverables without building foundations - was one of the strongest themes in the research. Fergus extended it at the webinar into something more generative.

His argument was that the technology industry's long obsession with removing friction is, at its core, bad design. The question isn't how to make systems frictionless. It's how to locate and preserve the correct friction - deliberate pauses and checkpoints, moments that force real engagement with a problem.

I think it's bad design. There should be the correct friction at every point - whether that's in how you analyse the media you're consuming, or in AI outputs, or in your own confidence based on built-up expertise - Fergus.

This resonates with a wider and counter intuitive design principle that Mark has often spoken about and learned from designer Louisa Heinrich at Fjord many years back. We should not actually aim for seamless. Seams define a product.

This applies to education, to management, to the design of tools and services. An organisation that strips out all friction doesn't produce faster thinkers, it risks producing people who have not been asked to think.

Sophie Wade extended the point from a different direction. Her observation - that LLMs are not neutral, because they are trained on a century of human content including fear, suffering and love - connects to this. An AI that detects a user's stress and simplifies its answer to provide reassurance isn't being helpful. It is removing friction in precisely the moment when friction would serve that person best.

Isolation is a design failure, not an inevitability

The isolation finding in the research described a structural shift: if teams shrink, culture deteriorates, eventually the social texture of work dissolves. Fernanda had put it with acutely: projects that once involved five people now involve two. You move faster but alone.

At the webinar, Sophie Wade reframed the problem - not to soften it, but to make it actionable:

We haven't designed work for what we actually have now. We went from factories to offices that looked like factories - and we still haven't designed for location-independent, human-centred work - Sophie Wade, work transformation strategist.

The implication is significant. Isolation isn't an inevitable consequence of AI - it's the consequence of deploying AI into work structures designed for a different era and leaving those structures unchanged.

Fernanda, who has spent the months since the original research actively trying to rebuild human connection in her workplace - more office time, volunteering, team rituals, an analogue diary - was cited by Sophie as an example of exactly the kind of self-directed redesign that needs to happen at institutional scale. What one person is doing alone and off the side of her desk is actually what organisations should be doing deliberately.

Two kinds of knowledge

One of the clearest new ideas to emerge from the webinar - developed between David and Sophie in the moment - is a distinction the research gestured toward but never made explicit.

There are two kinds of knowledge inside organisations. Knowledge that can be written down, systematised, and eventually instantiated in a large language model. And knowledge that cannot - empathy, the ability to inspire a team, relational intelligence, the feel for how a room is going. The second kind is, arguably, the more important. And the mechanisms for transferring it between people - mentorship, apprenticeship, proximity, Xerox engineers sharing what they knew over coffee (a famous case study from ‘The Social life of Information’) — are exactly the mechanisms that tech-driven efficiency pressures tend to cut first.

Your challenge is going to be to build a culture where AI enriches your critical thinking rather than degrades it. Everyone's going to have the magic button. The difference is still going to be people - doing the things that only people can do - David Mattin, co-founder, Full Moon.

This is the organisational version of Skyla's dilemma. Can you protect the conditions under which things the tool can't do still get learned?

Who owns the intention - and who provides the vision?

The research concluded with a call for vision. The young people we spoke to were not, fundamentally, asking about job numbers. They were asking about meaning, credibility and intent. Who owns the intention?

We asked the panel who they would trust to provide that vision. The three answers were instructive, and usefully different.

Fernanda said she would trust someone older, with accumulated wisdom and presence. Her grandmother's challenge about her smartwatch - you're not more conscious about your health, you're just outsourcing awareness to an app - drew nods. Perhaps we need someone with that quality of attention - an 80-year-old, she said, without irony.

Fergus said age was irrelevant and went for dialogue not deference. What matters is whether the person can explain their reasoning in terms he can interrogate and argue against.

Skyla said she would trust someone she connected with - not on interests, but on values. Someone - not an institution - whose moral framework felt legible to her.

These are three very different answers. What they share is that trust is earned, personal, and explicitly not conferred by authority or credential. Institutions that think they can claim the guidance role simply by publishing a policy are misreading what young people are actually asking for.

The AI super user finding - a complication worth noting

The isolation narrative, which runs through the research and is among its most consistent findings, was complicated by one data point from Amanda Schneider.

Gensler research on AI super users - the heaviest users of AI tools in workplace settings - found that these people actually spend less time working alone, not more. They freed up time that they reinvested in learning, socialising, and moving across tasks. They were measurably more mobile and more connected than lighter AI users.

This doesn't invalidate the isolation finding. But it suggests a more precise formulation: isolation may be the consequence of poor or unreflective AI implementation, in organisations that deploy the tools without redesigning the work. It is not an inevitable property of AI use itself.

The difference, as Amanda put it, is between a car with its wheels spinning off the ground and one with the wheels on the road.

Culture is the output

The most memorable line of the session - and the one that most cleanly captures something the research circled without landing - came from Amanda:

Culture is the outcome. It's not the input. Everyone wants the injection. Nobody wants the diet and exercise plan - Amanda Schneider, founder of ThinkLab, author of Work for What's Next.

The research found institutional abdication on a significant scale. This is a sin more of omission than commission. Governments, universities and employers are not giving young people meaningful guidance on navigating AI and work. The webinar surfaced why this is harder than it sounds. This is certainly something to think about, hard.

Building culture in the age of AI is not a communications problem or a policy problem. It is a sustained, intentional, trial-and-error process that looks like diet and exercise and takes the same amount of time. Every organisation wants the shortcut. There isn't one.

What comes next

The webinar generated more questions than it answered - which is, we think, the right outcome. Among the threads we want to return to:

The mid-career story. The research focused on young people entering the workforce. Sophie's point about flattening organisations - fewer middle managers, wider spans of control, more people being asked to do more with less - suggests that the 35–50 cohort may face a structural disruption that is distinct from, and in some ways sharper than, the entry-level story. This deserves its own treatment.

The generational comparison. Esther Stringer asked, in the chat, whether previous generations felt the same way at the same age. We don't have that data. It's the right question and we're interested in pursuing it - carefully, with the comparative evidence it would require to answer well.

The culture question. David named it live: culture in the age of AI is one of Full Moon's core ongoing themes. It's too large to resolve in a webinar. We'll be returning to it.

The vision question. Is it possible to conjure up a vision for the future of work that is engaging and motivating? To what extent is that down to the individual to grapple with - now?

 The original essay - AI & Work: How Young People Are Thinking About the Future - is available here.