Career what-if engine
It is a bundle of tasks. Some of them are about to change hands.
Not overnight. One task at a time, faster in some futures than others.
Into a real job you can grow into, with a plan that fits your week.
Career what-if engine
Oroboros shows which of your tasks AI may take over, under slow and fast futures, and which real job the rest can grow into.
The idea
A job is a bundle of tasks. AI does not delete jobs overnight. It eats some tasks inside them, and what is left over becomes a different job. Oroboros shows a person that process, on their own work.
We list your real tasks and show which ones AI is likely to absorb, using live public signals.
We match what is left to real occupations, or a blend of two, and show the gap to each.
You get an ordered learning plan with free courses, hours per step and a proof task for each.
"I don't need fifty job ideas. I need to know which parts of my job to bet on, and what to learn first."
How it works
Every number in the app is computed by code from public data. The same three formulas are printed next to every result.
mc = clamp( last 90 days ÷ past-year average, 0.8, 1.2 )
For each of ten AI capability areas we compare the last 90 days of arXiv papers, GitHub commits on reference projects and Hacker News hiring posts with that area's own past year. The cap keeps a noisy feed from taking over.
Modelling assumption, refreshed nightly.
At = min( 1, Et × mc × s )
Et is the task's exposure from published datasets. We keep the lowest and the highest estimate, so when datasets disagree you see a range, not a fake single number. s is the scenario: Slow 0.6, Gradual 1.0, Shockwave 1.6.
Consumed at A ≥ 0.6 · shrinking 0.35 to 0.6 · safe below 0.35. If the range crosses a line, the task is marked uncertain.
Rank = Fit × Resilience
Fit is how much of a role's core work you already do. Resilience is how much of that role survives the same scenario, so we never send you to a job that is being eaten too.
Composite roles are always two real, named occupations. We never coin a title.
Try it
Pick a synthetic persona or your own occupation. Everything runs in your browser. Nothing you type leaves this page.
Pick a persona, or search an occupation and tick the tasks you actually do. Set the hours each takes in a normal week.
How fast each AI capability is moving right now, from live public feeds. A reading above 1.0 means more activity than its own past year.
Modelling assumption.
Sources
Layers that change by release show a version. The live layer shows when it was fetched. If a feed fails, the site keeps the last good snapshot and says so.
We do not predict that, and nobody can honestly. We show which tasks inside your job are exposed, under three stated assumptions, and what you can build from the rest.
Tasks come from O*NET 31.0. Exposure comes from three published datasets, kept separate so their disagreement shows as a range. Momentum comes from arXiv, Hacker News, GitHub and job boards, refreshed every night. Every chart says where and when.
O*NET is the most detailed open task database there is, and it is US-based. Malaysian job titles map to it through the ISCO occupation codes that MASCO also uses. We flag this on every screen instead of hiding it.
No. The engine runs in your browser. The demo uses five synthetic people. Please do not paste a real CV with personal details.
The core uses no language model at all. Two optional jobs, reading pasted CV text and writing plain-language summaries, use Qwen2.5-7B-Instruct. Every role, task and course it mentions is checked against our database first, or dropped.
No. Oroboros is for the person doing the job. It does not rate people and must not be used for hiring or firing decisions.