The WholeElephant

That’s most of us with AI right now. We’ve each touched one part — a few prompt tricks, one good demo, one project that went wrong — and we call our part the whole animal.

Job AQ (jobaq.me) walks you around the whole elephant. It’s a free behavioural assessment of how well you actually work with AI — not what you know about it — built from scenarios drawn from your own job. Think of it as IQ, but for AI.

At the end, you get four things. A score out of 100. Your four dimensions, which show where you’re strong and where your blind spot is. Your AI persona. And a Skills Map showing where AI would pay off in your job.

This book explains what the assessment measures and what to do with the results.

Someone strong on all four can be handed AI-heavy work without worry. Someone strong on only three will fail in a predictable place — and the assessment tells you where.

Scenarios, not questions. Job AQ doesn’t ask you about any of this — it puts you inside the scenario. In one, you decide how much of a real task to hand over, and sometimes the right answer is more AI, not less. In another, you read a polished AI-written deliverable and mark what you’d check before signing off. In another, AI-assisted work has gone out under your name and turned out wrong, and you work through what happens next.

Why not another AI quiz. Most AI assessments measure recognition. Can you define hallucination? Which prompt is better? That’s awareness, the lowest level of skill — which is why people who can define hallucination still miss a fabricated benchmark in a real document. Job AQ measures behaviour instead. It also scores both directions of error the same way: handing too much to AI costs you, and so does refusing it for work it does well.

Sixteen signals. Each dimension breaks into four sub-skills. We cross-checked the sixteen against the AI-fluency frameworks already out there — the ones published by the big AI labs, government skills taxonomies, academic instruments. We cover most of what they cover, plus five we didn’t find in any of the ones we checked:

• Manage AI’s effort — knowing when the time AI saves you in drafting gets eaten up by the time you spend checking its work.
• Spot the irreversible — recognising decisions that are hard to undo once they spread downstream.
• Check the substance — noticing when output sounds confident but says nothing useful.
• Own the mistakes — raising AI-assisted errors quickly, and being clear about AI’s part in them.
• Own your voice — making sure what goes out reflects how confident you actually are, not AI’s default certainty.

One caveat. Each run has only about one question per sub-skill, so treat the sixteen as pointers to explore, not precise scores.

Twelve archetypes. Modes of working, not job titles. A “Senior Manager” title tells you seniority and your company’s naming habits. Not whether the week is spent building models, closing deals, or reviewing compliance.

The twelve come from large occupational datasets — over a thousand occupations, tens of thousands of task statements — cross-checked against skills taxonomies and AI-exposure research. Each archetype carries its own skills, each rated for how much AI can genuinely help. Those ratings come back later, in your Skills Map.

Most real jobs span more than one archetype. So you can pick a primary and a secondary, with a slider for how your time splits. A product-minded engineer may be a Builder-Strategist. A founder may be closer to a Leader-Closer.

AI also hurts each archetype differently. Take the Analyst. When AI writes a weak strategy memo, people may sense the bullshit and push back on it. But when AI makes up a number in an analysis, people may take it as fact and build decisions on it. And a made-up number looks exactly like a real one, so there is nothing odd on the page to warn anyone. That’s why your archetype’s scenarios are built around the specific ways AI causes problems in your kind of work, not generic ones. And the archetype changes what you see, never how you’re scored. An Analyst’s 74 and a Storyteller’s 74 mean the same thing.

Belief ahead of behaviour: overconfidence. You rate yourself a careful verifier. You missed the planted fabrication. Confident people get handed the high-stakes work where blind trust costs most. Which failures did you miss, and what would checking for them look like?

Behaviour ahead of belief: quiet competence. You call yourself a beginner, then delegate, verify, and push back like an expert. People in this quadrant undersell themselves in exactly the meetings where their judgment should count.

Where do you think you are? And where do you want to be? The assessment shows you the gap, and the report gives you a path to close it.

• Automator — high delegation, lighter scrutiny. You automate the repetitive. Your risk is the day the repetitive turns out to matter.
• Augmenter — high delegation, high scrutiny. AI as a thought partner. Hand over a lot, check a lot. Sustained high-trust AI work lives here.
• Judge — low delegation, high scrutiny. You evaluate before you commit. Deliberate and quality-first, but sometimes too slow to catch the opportunity.
• Observer — low delegation, lighter scrutiny so far. Watching and learning. Everyone starts somewhere.

Observer to Automator to Augmenter. Adoption first, judgment next. Or maybe you are the rarer Judge — scrutiny before adoption.

The score indicates how you work with AI. The persona shows how you relate to it. Two people can share a 70 but be different. The Automator should verify more. The Judge should hand more over.

It’s a snapshot. It comes from what you said about yourself. And stances change. Retake after a quarter and you may well land in a different quadrant.

• Sweet Spot — high relevance, high adoption. Keep going, go deeper.
• Opportunity — high relevance, low adoption. AI could change these tasks and you haven’t started.
• Low Priority — low relevance, low adoption. Correctly ignored. Leaving these alone is good judgment.
• Over-Invested — low relevance, high adoption. You’re forcing AI into work it doesn’t fit. Usually the verification overhead eats the drafting speed-up.

The map has two axes: how much AI could genuinely help with each task, and how much you use it on that task today. The relevance rating comes from the same occupational research as the archetypes, computed on the backend. You can’t nudge it.

The Opportunity corner is where to start: the highest-relevance task you’re not using AI on yet.

The blind men in our story are stage one: certain, wrong, and unaware of it. That’s where most of us sit with AI. And it’s the dangerous one, because at stage one, confidence tells you nothing.

An assessment has one job: move you from stage one to stage two. Before, you had a feeling about how you work with AI. After, you know which dimension, how far off, and how that compares to what you believed. That’s progress, even if the score stings a bit.

At the end of the quarter, question yourself: Did the situation come up? Did you do the thing? Keep the ones that worked, sharpen the ones that didn’t, and replace any whose situation never came up. Then retake — your second score tells you if the plan worked.

Almost all AI training is evaluated at level one, with a satisfaction survey at the end of the workshop. Some reaches level two, with a quiz. Behaviour — whether anyone actually works differently on a normal working day — almost never gets measured, because it’s hard to measure. Which is a pity, because that’s the level where the money and the risk sit.

Job AQ is a level-three measurement you can run in fifteen minutes per person. Before any training, it tells you which behaviour actually needs to change — a team weak on Evaluate AI needs verification practice, not another prompting workshop. After the training, a retake tells you if behaviour moved.

The modern version of the Kirkpatrick model adds one more useful idea: plan backwards. Start from the result you want (fewer unverified AI outputs reaching clients, faster turnaround without more errors), work back to the behaviours that produce it, and only then choose the training. Most organisations run this exactly in reverse — they buy the course first and hope for a result.

The measurement also shows you some things worth acting on. First, some people work with AI better than they rate themselves. They make good reviewers of other people’s AI-assisted work, but because they undersell themselves, nobody asks them. Second, teams often share the same weakness. One person weak on Evaluate AI is a gap. A whole team weak on it is a risk, because everyone assumes someone else is doing the checking.

Plan, week one. Take the assessment. Your weakest dimension is the behaviour to work on. The Opportunity corner of your Skills Map is the task to start using AI on. Write your two or three if-then plans from the last chapter. That’s the plan — just one behaviour, one task, three sentences.

Do, weeks two to twelve. Work as normal. Nothing new goes into your calendar. Use AI on the Opportunity task, with a light check. And follow your if-then plans whenever the situations come up.

Check, week thirteen. Retake. Fifteen minutes. The difference between the two scores tells you if your behaviour actually changed.

Act. Score moved? Pick the next dimension and go again. Score didn’t move? Your plans were probably too vague. Make the “if” more specific, make the “then” smaller, and run the same dimension again.

Same loop for a team, just with more people. Week one, everyone takes it. Week two, look at the results together and pick one shared weakness, one owner, one fix. Spend the quarter on it. Week thirteen, everyone retakes. Now you have a clear before and after.

Nothing here is about trying to be clever. It’s about being simple enough to repeat. And improve.