Data & method

ceiling c   = β from Eloundou et al. (1: an LLM alone could halve the task's time, 0.5: with extra tools, 0: no)
today   e₀  = c × usage intensity      usage intensity = Claude share of the task ÷ 90th-percentile task share,
                                      at least 0.2 for any task the Anthropic Economic Index publishes
in h years  e(h) = c − (c − e₀) × (1 − r)^h        r = the scenario's yearly adoption rate

Each task also gets an automation share: of the conversations about it, how many hand the work over (directive or feedback-loop use) rather than work alongside a person (iteration, learning, validation). That splits exposure into work AI is likely to automate and work it is likely to assist.

outlook = 50 + 20 × demand + 40 × assisted share − 55 × automated share + 5 if on MyCOL     (0–100)
demand  = (sector postings year-on-year − all postings year-on-year) ÷ 15%, capped at ±1
MyCOL   = the role is on Malaysia's Critical Occupations List (TalentCorp), i.e. hard to fill locally

Labels compare each role with all 30 roles SkillTide tracks: at least 0.6 standard deviations above the average is rising, 0.6 below is under pressure, and a role where AI touches at least 45% of the work is transforming. Demand is measured against all postings so a hiring slump that hits every sector does not make every role look bad.

Hiring data comes by Indeed sector, not by job title, so roles that share a sector share a demand value. In the latest snapshot (2026-10-16), 12 roles drawing on 3 sectors are at the cap; between them, outlooks differ only through the AI signals and the MyCOL bonus.

skill score = 50 + 25 × momentum + 30 × role signal
  momentum    = change in the share of Hacker News hiring posts mentioning the skill (last 6 months vs the 6 before; ±50% → ±1)
                AI skills average this with the change in the share of Indeed postings that mention AI
  role signal = how far the outlook of the roles needing the skill sits from the average role (±12 points → ±1)

match = (w_skills × skills fit + w_outlook × outlook + w_sdg × SDG fit) ÷ (w_skills + w_outlook + w_sdg)
  skills fit = importance-weighted share of the role's skills you have at the expected level
               (another programming language you know counts as 70% of the way)
  SDG fit    = share of your chosen goals the role contributes to (switched off if you pick none)

Role requirements come from O*NET: technologies O*NET flags as in demand for that occupation, rated skills and knowledge, and skills named in its task statements. Each requirement keeps that evidence. The pathway addresses the highest-priority technical gaps first, choosing courses that cover the most priority per hour of study within your budget, schedules them by your weekly hours, and ends with a ~20-hour portfolio project built on a real task from the role that AI assists rather than replaces.