For a while we've been watching what people do with their results. Most read the score, scroll through the task list, and leave. A small number share the page. Almost nobody shares the number itself.
This made sense when we thought about it. A percentage exposure score is information about you, not something you want to lead with. "I got 54% automation risk" isn't a conversation starter — it's a vulnerability disclosure. Nobody voluntarily posts that on LinkedIn the way they'd share a personality type or a quiz result.
The underlying problem is that a score is an endpoint. It doesn't invite anyone in. It doesn't say anything about who you are, how you work, or what makes your job hard to replace. It just measures a single axis of something people are already anxious about.
So we went back to what the model already knows. Every assessment doesn't just produce an exposure number — it produces two independent dimensions: how knowledge-intensive the work is (substrate depth), and how much the work depends on social and relational context (social dependency). These two dimensions have always driven the gap between our naive score and the effective score. The Eloundou et al. study found a similar pattern — raw task-type exposure and adjusted exposure diverge significantly depending on the human dimensions of work. We've been displaying only the outcome of this calculation, not the structure underneath it.
The four archetypes come directly from that structure.
The Strategist has both thick knowledge substrate and high social dependency. The work requires deep expertise and it happens inside complex human systems — organisations, relationships, stakeholder environments where trust and judgment are inseparable. Senior managers, consultants, therapists, executives. The exposure score tends to be low, but more importantly, the reason it's low is meaningful: this work is hard to reach precisely because it sits at the intersection of two things AI has the least purchase on.
The Craftsperson has deep knowledge substrate but lower social dependency. The work is mastery-driven — specialised, skilled, often hands-on or deeply technical — but not primarily relational. Surgeons, architects, engineers, researchers. The protection comes from depth, not from social embeddedness. AI can accelerate the research and the drafting; the judgment at the core of the work is harder to compress.
The Connector has high social dependency but a thinner knowledge substrate. The work is fundamentally relational — coordination, facilitation, care, communication. The value delivered is trust, presence, and the ability to read and navigate people. AI can simulate these things in controlled contexts but not in the unpredictable, emotionally loaded environments where Connectors operate.
The Optimizer has a thinner substrate and lower social dependency. These roles tend to sit closest to the automation zone — structured tasks, digital workflows, well-defined processes. This isn't a comfortable position, and we didn't want to pretend otherwise. But we also didn't want to frame it only as risk. The people who thrive in this quadrant won't be the ones who wait. They'll be the ones who develop fluency with AI tooling early, redeploy their time toward higher-leverage work, and move before the transition happens to them.
One thing we noticed quickly when we started testing this: people engage very differently with a type than with a number. "I'm The Strategist" is something you might say in a conversation. "My automation exposure is 31%" is not. The type carries meaning about how you work and why that's defensible — which is the thing people actually want to know, even if they came to the site to get a score.
We're not trying to make people feel better about their situation. The exposure data is still there, still honest, still broken down task by task. The archetype sits above it — not as a softener, but as a frame. A score tells you where you are. A type starts to tell you why, and what to do about it.
Every assessment now shows your type on the results page. If you've already mapped your job, the type was derived from the same data you submitted — nothing new to fill in.