TaskFolio: Seeing Which Parts of Your Job AI Will Actually Affect
I built a tool that breaks down AI's impact on Australian jobs at the task level — not just 'your job is 40% automatable' but exactly which tasks, when, and how.
Everyone talks about AI replacing jobs. The numbers are always scary and abstract: “47% of jobs at risk of automation.” But that framing is wrong. Jobs don’t get automated wholesale — tasks do.
A financial analyst’s job involves data entry (highly automatable), building relationships with clients (not automatable), and interpreting ambiguous market signals (partially automatable, with human oversight). Saying the job is “47% at risk” obscures more than it reveals.
The Spark
The idea crystallized while listening to Paula Pant’s interview with Dr. Ben Zweig, CEO of Revelio Labs, on the Afford Anything podcast. Zweig’s core insight stuck with me: jobs aren’t indivisible units that either exist or don’t — they’re bundles of tasks that are constantly evolving. The real question isn’t whether AI will replace your job, but which tasks within your job will be automated.
Then Andrej Karpathy released his US Job Market Visualizer — a beautiful treemap showing AI exposure by occupation. @ychua built an Australian version using ANZSCO codes and LLM-powered scoring. Both were occupation-level views: “Software Developer: 9/10.”
I wanted to go one level deeper: not just which jobs, but which tasks within those jobs, and when.
That’s TaskFolio.
The Data Behind It
TaskFolio combines three data sources:
1. Anthropic’s Economic Index (January 2026)
Anthropic analyzed 1 million real Claude conversations to see what people actually use AI for professionally. This isn’t a survey or prediction — it’s empirical data on current AI usage patterns. They matched conversations to occupational task categories, giving us ground truth on which tasks people are already delegating to AI.
2. O*NET Task Decomposition
The U.S. Department of Labor’s O*NET database breaks every occupation into granular tasks. A “Software Developer” isn’t a monolith — it’s 15-20 distinct activities like “analyze user requirements,” “debug code,” “collaborate with stakeholders,” and “write documentation.” Each task has different AI exposure.
3. Jobs and Skills Australia Projections
For Australian context, I integrated JSA’s official 10-year employment projections (May 2025 → May 2035). This adds the labor market reality: is this occupation growing, shrinking, or stable? High AI exposure in a declining field feels different than high AI exposure in a booming one.
What You Can See
The homepage shows a treemap of 361 Australian occupations, sized by workforce and colored by AI exposure:
- Red: High exposure (>60% of tasks affected)
- Yellow: Medium exposure (30-60%)
- Green: Lower exposure (<30%)
Click any occupation to see the task breakdown. Each task shows:
- AI exposure score: How much of this task AI can currently handle
- Impact type: Will AI replace it, augment it, or have no effect?
- Timeframe: Is this happening now, in 2-5 years, or 5-10 years?
- Success rate: How often does AI currently succeed at this task?
The occupation page also shows a Future-Proof Index — a single score combining AI exposure, pay risk (lower pay = easier to automate economically), and employment outlook (growing fields can absorb AI impact better).
The “Half-Life” Metric
One thing I added that I haven’t seen elsewhere: a Half-Life estimate. Based on timeframe distributions, when will 50% of your tasks be AI-affected?
An occupation with half-life of 2 years is qualitatively different from one with half-life of 8 years, even if both have the same total exposure. The first needs immediate adaptation; the second has runway.
What I Learned Building It
Task-level data is sparse. Anthropic’s index doesn’t cover every O*NET task — about 40% of occupations needed synthetic task decomposition using the patterns from their covered occupations. The methodology documentation explains the confidence levels.
Timeframe estimates are the weakest link. Anthropic’s data shows what’s happening now. Extending to 2-5 year and 5-10 year horizons required extrapolation from technology adoption curves and regulatory considerations. These are informed guesses, not predictions.
Employment projections add crucial context. A data entry clerk with 85% AI exposure and -5% employment growth is in trouble. A registered nurse with 45% AI exposure and +25% employment growth is probably fine — the augmented tasks free up time for the growing demand.
Try It
Live site: ai-job-exposure.setiyaputra.me
Source code: github.com/suryast/task-folio (MIT licensed)
The data pipeline, methodology, and scoring formulas are all documented in the repo. If you work with labor market data or AI impact research, I’d love feedback on the approach.
The tool covers 361 ANZSCO occupations representing 14.4 million Australian workers. It’s free, no login required, and the source is open. The only proprietary piece is the curated dataset — the synthesis of Anthropic, O*NET, and JSA data into a unified task-level model.
Built with Next.js 16, Cloudflare Workers (D1 + Pages), and D3.js. The neobrutal UI is intentionally bold — job automation data shouldn’t hide behind soft gradients.