Wildcard track
Who are you with AI?
A three minute, game-like run that tells you what kind of AI worker you are today, and then shows you where to explore next in your own job. No scores, no ranking, no jargon, and it never tells you which product to buy.
Password: explore. It is a doormat, not a lock: nothing private lives behind it.
The problem we picked up
The evening kept circling one thing: everyone is told AI matters for their job, and almost nobody is told what that means for their job. The advice on offer is either a product demo or a maturity assessment that hands you a number and a verdict.
Both fail the same person: the colleague who has never opened a chat box, who suspects the answer is "you are behind", and who therefore never starts. Meanwhile the people already using AI get generic tips they have read ten times.
So we built the opposite. It asks about your work before it says anything about AI, it never ranks you against anybody, and the output is your own ideas, played back to you as a place to start.
How it works
Phase 1 · Who are you with AI?
Four playful stops, one per screen, with visible progress. Before they start you introduce yourself in two or three sentences, exactly as you would at an event. From that, the tool guesses the five tasks that fill your week, shows them to you, and lets you drag them into the right order or throw out the ones that are not your job. Everything after that is written around your actual work.
-
Quick instincts
A deck of one line situations from your week. Swipe yes or no. No right answers, no reading.
-
Pack your bag
An impossible week is coming. Pick the three moves you would actually make. The combination is what counts.
-
The dial
One slider, from "all me" to "it drafts, I send", set against a real piece of work you named yourself.
-
The wish
The first thing you would hand over if you could. Free text, deliberately never judged, because a fake verdict on your own words would be worse than none.
You then meet one of five characters that describes how you work with AI, plus a stage that says how far along you are. Both are computed separately, so no character is better than another, and every stage name points forward.
The Explorer
Tries everything, hands over boldly. Brilliant at starting, sometimes too trusting of what comes back.
The Artisan
Briefs precisely, refines until it is right. Gets great output, and sometimes polishes what should have been delegated.
Detective · Guardian · Scout
The one who checks everything, the one who thinks about rules and risk, and the one still surveying from the ridge.
Phase 2 · What could you do with it?
Instead of a list of tips, your result opens as a board game map: regions you can wander, each holding small concrete experiments, with one region highlighted as "closest to what you described, start here". Every suggestion is tool agnostic. The concreteness sits in the verb, not in a brand name: take the last report you wrote, paste it into whichever assistant your company gives you, ask for five bullet points.
What is real today, and where the tape is
Real and working
- All of Phase 1, end to end, in the browser: intro, task ranking, four stops, character reveal, stage, and a plain language strength and weakness.
- Live personalisation through two Claude calls. The first reads your intro and guesses your week, using web search so that "I am a controller" still lands somewhere real. The second rewrites the station texts around your confirmed tasks.
- The five characters as animated vector figures, and a markdown export so you can take your result with you.
- Silent fallback at every step. No key, no network, a timeout or bad output, and you get the static version instead of an error.
Honestly stubbed
- Phase 2 is one hand crafted example map for a project coordinator called Alex. It is labelled as an example in the product. There is no live generation yet, and we say so on stage.
- The scoring thresholds are hand tuned. Real people have not run through it enough to trust them.
- Desktop only for now, which is awkward, because a shared link mostly opens on a phone.
- Entry is a single shared password checked in the browser. Good enough for an internal pilot, nothing more.
The rules we gave ourselves
The measurement stays hidden
Under the surface we score five dimensions, four of them from Anthropic's public AI Fluency framework plus one for how much AI you have actually touched. The user never sees an axis, a bar chart or a number. They see a character and a stage.
No ranking, no verdict
No beginner label, no percentage, no percentile, no comparison with other people. At the low end a percentile produces the single most discouraging sentence this product could show, aimed at exactly the audience it was built for.
No product names
We never recommend a tool. Companies have different stacks and rules, and a recommendation you are not allowed to follow is worse than none.
No use case bias
We do not suggest what AI is good for before you have described your own work. People should arrive at their own ideas, which is why the intro comes first.
How it is built
Vanilla HTML, CSS and JavaScript with ES modules. No framework, no build step, no dependencies, deployed as plain files on Cloudflare Pages. Nothing in it needs a framework, and zero tooling means the demo cannot break at build time.
Two Claude Sonnet calls run straight from the browser and touch display text only. The scoring engine is never model driven: one engine, one source of truth, so a rewritten question can never quietly change what an answer means. Your answers stay in local storage. Only your short intro leaves the browser, once, and the page says so before you type.
The whole thing, concept to live site, was built in one day with Claude Code, including the four feedback rounds that rebuilt Phase 1 from a questionnaire into minigames.
Would we keep going, and what would it need
Yes, in this order:
- Phase 2 for real. A curated library of 30 to 60 exploration cards, assembled per person by one model call. Curated on purpose: free generation is great four times out of five and embarrassing the fifth, and the fifth happens on stage.
- Real test runs. Enough people through Phase 1 that the character you get feels uncomfortably accurate, which is the only quality bar that matters here.
- A phone layout, because sharing a link is the entire distribution plan.
- Grown up data handling before anyone types about their actual job at scale: a visible confidentiality warning, a retention period, a data processing agreement, and no training on the input.
- A public name. The working title is a placeholder and a cartoon trademark, so it cannot ship.