The Four Archetypes of the AI-Split Workforce

Most AI adoption programmes are built for a population of two: people who are on board, and people who need convincing. Noam Segal and Lenny Rachitsky’s annual sentiment survey, now in its second year and covering roughly 6,000 tech workers, found four. The two biggest groups were never resisting in the first place, which means most adoption effort is aimed at the wrong 24% of the room.

Four groups, not two

The survey sorted respondents into four emotional postures toward AI at work:

• The Energized (41%) report feeling excited, curious, and hopeful, the only group with a genuinely positive outlook on where their field is heading. One respondent put it simply: “product has become fun again, you become an explorer.”
• The Conflicted (35%) hold positive and negative feelings at once, curious and overwhelmed and tired, all in the same week. One described it as “simultaneously having the most fun and the most uncertainty.”
• The Disoriented (12%) aren’t opposed to AI, they’re losing their footing as their own role shifts faster than they can adapt to it. Their own comparison: “like farmers on the cusp of the industrial revolution.”
• The Resentful (12%) all selected the same feeling: pressured to use AI. Not that the tool is bad, that the ask stopped being a choice. “We used to adopt tech for excitement, now it’s use AI or lose your job.”

The 76% nobody’s message is for

Add the Energized and the Conflicted together and you get 76% of the workforce, and neither group is what an adoption programme is built to move. The Energized don’t need convincing, they need someone to stop repeating a pitch they already bought. The Conflicted don’t need convincing either. They need something adoption programmes rarely offer: permission to hold both feelings at once, instead of a message that implicitly asks them to resolve the ambivalence into enthusiasm.

That gap shows up in the numbers around it. Burnout across the full sample rose from 44.7% to 55.7% in a single year, but 82% of respondents still say AI makes them measurably more productive, in the same breath as describing the trade-off the way one respondent did: “I’m amplified, but my brain is rotting, and my work feels worse.” That is not the profile of a population that needs persuading. It is the profile of a population managing a real trade-off nobody is helping them name, let alone navigate.

I’ve sat across from people in exactly that state: genuinely delighted by what they had just shipped in an afternoon, and visibly unsettled by how little they trusted their own judgment on whether it was actually right. Both things were true in the same conversation. Neither reassurance, it’s fine, trust the process, nor another training session addressed what they actually needed, which was permission to say both things out loud without either one cancelling the other.

The two fears the same survey found dominating this group aren’t about jobs disappearing either. 51% worry about being expected to do more for the same pay. 46% worry about getting trapped in a pace that can’t be sustained. Neither fear is about AI taking the work away. Both are about AI making the same work heavier, indefinitely, with no one renegotiating the terms.

Where the actual resistance lives

The Resentful and the Disoriented are the 24% an adoption programme might actually be built for, and even there, the standard playbook misdiagnoses both. I’ve written before about the identity threat driving AI resistance: people who quietly hang back are usually protecting the professional value they built their identity around, not rejecting a tool. That mechanism explains the Resentful group precisely. Every one of them chose the same word, pressured, because the framing shifted from an invitation to a condition of employment. Reassurance about the tool doesn’t touch that. Only removing the coercion does.

The Disoriented aren’t protecting an identity. They are living through one dissolving faster than they can build the next one, and what they describe wanting isn’t reassurance or less pressure, it’s a map: what does competence look like now, concretely, in the role I’m actually doing next quarter.

One message, four audiences

None of this is solved by a better rollout email. A single adoption narrative, however well written, is being asked to do four incompatible jobs at once: stop repeating a sale to people who already bought, sit with people’s ambivalence instead of resolving it for them, remove coercion rather than explain the tool again, and hand out a map instead of a pep talk. Treating a genuinely differentiated population as one audience is the same structural avoidance that shows up everywhere else leadership prefers one clean initiative to the harder work of actually looking at who is in the room.

The tech workforce didn’t split into believers and skeptics. It split into four groups having four different experiences of the same change, and a programme built for two of them was never going to reach the other two, no matter how many times it got repeated louder.