AI-Skill Demand: Indonesia vs Global Remote — Data Notes

Key Figures

Generated directly from the published aggregate CSV: 9,194 postings with parsed skill tags, taxonomy v2-2026-06.

AI-skill presence by dimension

0.9% of skilled postings name at least one AI-engineering skill under the strict v2 taxonomy. Per-dimension shares:

Top 15 skills by demand

Pre-Registration

Hypotheses, operational definitions, exclusion rules, and statistical methods were fixed and timestamped before any results were computed — so the analysis could not be tuned to a desired outcome. This is the falsifiability anchor v1 lacked.

Six hypotheses were pre-declared (H1–H6). Each maps to a two-proportion z-test at α = 0.05. Results in the article report which hypotheses were supported and which were not.

Dataset Overview

Classifier Architecture

Taxonomy-first approach

The classifier is taxonomy-first: deterministic regex banks do the bulk of the work. An LLM pass fires only for postings where signals are absent or conflicting for task_altitude and seniority — the two dimensions where short or ambiguous titles can leave the regex banks without a clear signal. Tool/language/AI-skill counts are pure presence matches and never need LLM resolution.

The taxonomy version v2-2026-06 is recorded on every classified row, so re-running the same analysis over the frozen dataset reproduces every published figure exactly.

AI-skill taxonomy (six dimensions)

Classifier accuracy disclosure

Taxonomy version: v2-2026-06
Gold-set size: 200 stratified, hand-labeled rows (100 Indonesia + 100 global; balanced across seniority, task altitude, and AI-skill presence)
Acceptance gate: overall macro-F1 ≥ 0.80 on the deterministic path, measured against the gold set; below that threshold the taxonomy is revised before any figures are published.
Achieved: overall macro-F1 0.913 — AI-skill presence F1 0.992, task-altitude macro-F1 0.862, seniority macro-F1 0.886 (mean of the three).
LLM-path accuracy: measured offline against the same gold set; the share of rows resolved by the LLM pass is reported alongside the figures.

Statistical Methods

Every proportion reported in the article carries a Wilson 95% confidence interval. The Wilson interval is appropriate for proportions near 0 or 1 where the normal approximation is poor. It is computed directly from the counts — no external statistics package required.

Statistical tests use a two-proportion z-test (two-tailed, α = 0.05). Cells with n < 30 are flagged "directional only" and excluded from hypothesis conclusions. Results are observational — no causal claims.

Exclusion Rules

• Blue-collar: physical-labour roles, identified by title patterns (driver, chef, security guard, cleaning, waiter).
• Non-IT titles: Sales executive, cashier, accounting, HR generalist, content writer, and similar non-technical roles identified by title patterns.
• Empty titles:Rows with title length < 2.
• Salary: Excluded from all analysis. Disclosure asymmetry (Indonesia rarely publishes salary ranges; global sources vary) would confound any cross-market comparison.

Reproducibility

The method is described here in full so it can be replicated independently. The pipeline has four stages:

1. Assemble the corpus. A global-remote slice from the live job corpus, plus a one-time Indonesia-local listing scrape, both snapshotted to a frozen dataset for the analysis window.
2. Apply the exclusion rules. Drop blue-collar, non-IT, and empty-title rows (see the rules below); salary is excluded entirely.
3. Classify each posting. Deterministic taxonomy regex banks resolve tool, language, and AI-skill presence; an LLM pass resolves only the ambiguous task_altitude and seniority dimensions.
4. Aggregate. Compute per-dimension counts and shares with Wilson 95% confidence intervals — the figures in the aggregate CSV.

The published download is an aggregate CSV (counts and shares) drawn from our own job corpus; we do not redistribute the per-row scraped listings, which originate from third-party boards under their terms of use. With the taxonomy and exclusion rules described on this page plus a comparable corpus, the headline figures are reproducible end to end.

Limitations

• Non-probability sample. Indonesia-local data comes from listing-page scrapes, not a random sample of all Indonesian IT employers.
• Listing ≠ hire. A posting reflects stated demand, not actual hiring outcomes.
• Description quality varies. Indonesia-local listings are often shorter; regex classifiers may under-detect AI skills in brief postings (recall bias against Indonesia — conservative for H1).
• Single point in time. Snapshot from June 2026; AI tool adoption is changing rapidly.
• Classifier errors. Overall macro-F1 is 0.913 against the gold set, so a minority of rows are still mislabelled at the individual-row level — concentrated in the weaker classes (data-analytics altitude, F1 0.667; mid seniority, F1 0.750). Aggregate proportions are more reliable than individual classifications.

Version History

Questions or corrections: contact us. Methodology, classifier, and pre-registration are documented above; the published figures regenerate from the aggregate CSV.