Working Through AI Change

Summary

Most research on AI and work asks people what they think. We asked people what they were carrying — at the moment they signed up to spend an evening working on it.

Across … workshops between …, … founders, managers, and professionals answered a few optional questions when they registered: what is changing in your work, what is hardest about it, what do you want to walk out with. Nobody was being surveyed. They were asking for help, in their own words, before the room opened.

Five patterns hold across every room. The bottleneck is not AI information but a decision process that works under uncertainty. Overwhelm is decision load, not task load. Managers carry a distinctly relational burden, and they are overloaded rather than disengaged. Leaders ask for strategy while their teams ask for communication — the same gap seen from opposite sides. And the dominant emotion is not resistance. It is ambivalence: excited and overwhelmed at once.

Each pattern is stated with its number, its sample, and the external research it confirms, complicates, or contradicts. Each will move as new waves land. This paper is versioned so you can see how.

Where this comes from

I run workshops for people living through change at work — founders under pressure, managers whose teams are overwhelmed, people whose roles are shifting under them as AI arrives. Every registration form asks a few optional questions. For a long time I read them the way a facilitator does: to know who is coming. This spring I started reading them the way a researcher does, and something became hard to ignore. Across very different rooms, people were describing the same few things.

This paper is what those answers say when you put them side by side. The sample is what it is: people who chose to sign up for a workshop with "overwhelm", "role change", or "AI" in the title, mostly in and around San Francisco, mostly early-stage founders and the people who work with them. These are needs and intentions, not outcomes. I say more about the limits at the end and in the methodology. But inside those limits, this is evidence almost nobody else has: what people say when they are not being surveyed.

1. The bottleneck is a decision process, not AI information

The first thing I noticed is what people did not ask for. Almost nobody said they needed to understand AI better. What they said, over and over, was some version of I can't tell what to aim at.

"Nailing down a clear direction." — Founder

"What do I do now? Where do I go? How do I proceed with this AI startup I've already committed to?" — Founder

"How stay clear and focused with actionables in an ever shifting landscape of uncertainty." — Executive

For founders it reads as which strategy, which idea, which direction. For employees and people in transition it reads as what happens to my role and what should I build next. It is the one theme that unites everyone in the dataset, and it is highest — 83 percent — among people who describe themselves as in career or role transition.

I don't think this is a knowledge gap. Most of these people are unusually well informed about AI; many are building with it. What they lack is a way to make decisions when the ground keeps moving: what to weigh, what to let go of, when a decision is good enough. Microsoft's 2026 Work Trend Index found something adjacent from the other direction — the organizational conditions around a person predict AI value about twice as strongly as the person's own skills. The lever is not more information in the individual. It is the process around them.

2. Overwhelm is decision load, not task load

Overwhelm was the word on most of the event listings, so I expected people to describe exhaustion. Mostly they didn't. They described fragmentation.

"Because I wear different hats, my priorities are constantly shifting, and it can be hard to know what needs my attention first." — Operations lead

"Task management. I need to get out of this mode of doing tasks into thinking strategically 3 months ahead. And then going about doing things patiently." — Founder

"I want to be sharper at filtering real signal from noise, knowing what actually matters this week vs. what just feels urgent." — Founder

Notice the shape of that. It isn't I have too many hours of work. It is I have too many simultaneous open questions and no filter. Founders in our data barely name "decision fatigue" when it is offered as a phrase — five percent — but a third of the same kind of people pick "moving fast without creating chaos". The problem is pace and load, and they call it pace, not tiredness.

This matters because it explains why productivity advice bounces off this audience. A better calendar does not reduce the number of things you have to decide. MIT Sloan Management Review's work on employees "drowning in change" frames change fatigue as an organizational-load problem rather than a resilience problem, and that matches what we see. So does something the field only named this month: a study of six thousand digital workers found people spending roughly six and a half hours a week checking, correcting, and re-running AI's work — supervision labor that appears on no dashboard. Delegating to AI is not subtraction. It relocates work into judgment and checking. That is decision load, and it is what our rooms are already describing.

3. Managers are the fulcrum — overloaded and silent, not disengaged

Gallup's 2026 global report says manager engagement is falling roughly three times faster than employees'. That number worried me, because it is easy to read as managers are checking out. Our data says something different.

Managers are the only large group in our dataset where a majority name relational pain — trust, resistance, communication — as the hardest part. And when I look at the identical form filled in by founders and by everyone else, the people who report to leaders are two to four times more likely to name lack of trust, communication breakdown, and misalignment between leaders and teams. SHRM found the same gap from a different angle: about half of individual contributors trust leadership on AI, against four in five directors.

"Some of my coworkers are AI savvy and some like the pencil and paper. How can I involve my coworkers in accepting new thinking?" — Product IC

"Balancing … protecting an existing business while also building a new one — all while trying to lead an organization through a transformation to operate differently in the AI native era. I'm new in this role!" — Executive, new to the role

What I see in the room is not disengagement. It is people holding more decisions, more communication, and more of other people's fear than the role was designed for — and going quiet, because there is no obvious place to put it. Gallup also found that employees whose manager champions AI are almost nine times more likely to say AI is transforming their work. The manager is the transmission line. If the line is overloaded, nothing gets through.

The field caught up with this in August. Meta, which had pushed managers to fifty or more direct reports with AI coaching tools as the support, reversed course to about twenty, and its CTO wrote in an internal memo, reported by WIRED, that leadership had done an "atrocious" job explaining the vision. AI handled scheduling and status. It could not do trust, development, and someone actually knowing you. That is the relational layer our managers are describing, and it turned out to be load-bearing.

4. The leader–team translation gap

This is the finding I keep coming back to, and I hold it loosely because it comes from one form so far.

Founders and executives, asked what they want, say strategy — "a clearer AI strategy", "knowing what strategy to choose." Individual contributors, managers, and people in transition, on the same form, say clearer communication and better leadership alignment. Same room, opposite side of the same gap. The leader cannot decide. The team cannot hear a decision.

I've been noticing how often programs serve only one side. Leadership offsites help leaders decide and leave the team's experience untouched. Change communication helps teams hear and leaves the leader's indecision untouched. If the gap is real, the work is translation — the space between I've decided and we understand. I don't completely know yet what that work looks like at scale. But it is what our data says is missing.

5. Ambivalence, not resistance

Most change management starts from resistance: people don't want this, help them accept it. When we asked people to pick two emotions, that is not what they picked. The pattern was one activating emotion and one heavy one, together. Curious and uncertain. Energized and anxious. Excited and overwhelmed. Only about three in ten chose exclusively heavy feelings.

I recognize this in myself. I am building with these tools and I am unsettled by them, in the same afternoon. I don't think that is a contradiction to resolve. It is information about what people are actually holding — and it means programs framed purely as "coping with overwhelm" under-serve the energized half, and programs framed purely as "seizing AI" skip the heavy half. The most-registered events in our record had titles that held both.

6. What people actually ask for

Asked what would help, people wanted a practical next step and peer conversation in almost equal measure — and the free text was clear that the peer part must not feel like networking. "A less performative setting." Space to process uncertainty with others who are in it.

The external literature is thin here, which is part of why I think it matters. Nobody is publishing that leaders in the AI transition are lonely. But between founders naming loneliness at the top, "lonely" showing up unprompted in the emotion picks, and community language dominating what drew people to the smaller circles, it is in our data. Difficult change becomes bearable when it is witnessed. That is not a warm-up before the real content. In this audience it is the content.

7. What this means for leading AI change

I want to be careful here. Everything above is what the data says. This section is what I infer from it, and I'd rather be useful than certain.

Lead with the decision, not the tool. The unifying pain is I can't tell what to aim at. Any AI enablement effort that starts with training and ends with adoption metrics has skipped the part people are stuck on: a way to decide under moving conditions.

Design for both sides of the gap. Put leaders and their teams in the same room and work the translation between deciding and understanding. A program that helps only leaders decide leaves the team's pain — trust, communication, alignment — untouched, and that pain is what the data says is growing.

Treat the manager as the unit of change. Not as a channel for announcements but as the person carrying the most decisions and the most unspoken fear. Reduce their load before asking them to carry more.

Name the invisible work. Delegating to AI creates a second job — checking, correcting, deciding what to trust. Left unnamed it shows up as exhaustion and quiet corner-cutting. Named, it can be designed and budgeted for.

Hold both emotions. Say excited and overwhelmed in the same sentence. People will recognize themselves.

Make peer processing structural. Not a networking break. Structured, non-performative time with people in the same situation — and say so, because it is what people are asking for.

8. Limits, and what we will measure next

These are registration-time answers from a self-selected, San Francisco–heavy audience: evidence of what people bring to us, not base rates for the workforce. Roughly three quarters of the founders are pre-revenue. The questionnaire varied across early events, so cross-event counts use a conservative theme mapping described in the methodology. Where I have compared founders to everyone else on the identical form, I've said so.

Two things will change this paper. From now on every event carries the same five core questions, so waves are directly comparable. And a short post-event pulse will let us pair what people came with against what shifted — the difference between knowing what people worry about and knowing what helps. Two open questions we have not tested in our own data: whether the people adopting AI fastest report the most uncertainty, as Anthropic's 81,000-person study found; and whether the identity concern concentrates in early-career respondents. Both are on the next forms.

If you are living any of this — as a founder, a manager, or someone whose role is changing — I'd like to hear what your version of it is. That is how the next edition gets better.

Cite as

Imani, N. (2026). Working Through AI Change: what 363 founders, managers, and professionals tell us before anyone surveys them (v1.0). InsightsOut. https://insightsout.work/research/working-through-ai-change.html

Sources

• Gallup — State of the Global Workplace 2026
• Microsoft — 2026 Work Trend Index: Agents, human agency, and opportunity
• SHRM — Navigating AI in the Workplace, 2026
• Anthropic — What 81,000 people told us about the economics of AI (April 2026)
• MIT Sloan Management Review — When Employees Are Drowning in Change
• WIRED — reporting on Meta's internal memo on manager span of control (August 2026)
• Study of ~6,000 digital workers on AI supervision labor ("botsitting"), academic co-authors from Emory, Notre Dame, UC Berkeley, and UCL (August 2026)
• InsightsOut — Workshop Questionnaire Themed Findings, August 2026 (internal; aggregated and anonymized)