AI moved into the job ad. In Indonesia, the ads have not moved.

The question

Is AI changing what employers ask for, and is Indonesia behind or measured differently?

The usual answer counts "AI jobs" by title. That misses most of the change, because employers add AI to existing jobs far more often than they create AI jobs. It also mixes up countries with data sources: a country's numbers depend heavily on which job sites were read. We separate those effects before drawing conclusions.

Executive insight

Five findings

1. AI arrives as a requirement inside existing jobs.

   Global tech employers now name AI tools in 19.6% to 38.1% of ads in every job family, yet only 5% of ads are AI roles by title. Counting AI jobs undercounts AI demand several times over.

   FACT High confidence Evidence

2. Indonesian ads show no AI movement, on any board.

   The share of Indonesian ads mentioning AI went from 1.9% to 1.7%. The sample could detect a change of 0.31 points. Global engineering ads went from 3.4% to 23% over the same years.

   FACT High confidence Evidence

3. The gap is also in how ads are written.

   Global 2026 ads ask for ownership in 29.1% of cases; Indonesian ads in 5.1%. Indonesian ads ask for a degree in 48.9% of cases, against 10.3% globally. Indonesian ads describe credentials and tasks; global ads describe outcomes.

   FACT ANALYSIS Moderate confidence Evidence

4. The fall in Indonesian developer ads is a board story, and it is not AI.

   Developer titles fell from 11.3% to 7.1% of archived ads, but almost all of it happened on Glints and loker.id. On Kalibrr the share held at 16.2%. Glints filled the space with retail, hospitality and teaching ads. AI roles did not grow to replace developers.

   FACT ANALYSIS Moderate confidence Evidence

5. Employers now hire for range and results, not a list of tools.

   In global tech ads, large language models went from 0% to 10.4% of posts and ownership wording from 3.6% to 9.8%, while Java and Ruby more than halved. "Founding engineer" and "AI engineer" titles rose; "web developer" and "front-end developer" all but disappeared.

   FACT ANALYSIS Moderate confidence Evidence

One more result is about method: cross-country "share of tech jobs" mostly ranks the job sites, not the countries. Inside Indonesia alone, tech share runs from 2.8% on Glints to 48.4% on Careerjet. Evidence

How the system works

What sits between an employer's needs and the number we count

A share of job ads is the end of a chain. Read each row left to right: a force we cannot see directly, how it shows up in an ad, and the number we count.

1. Force · inferredShows up asWe count · measured
2. AI tools become cheap and useful at workEmployers add AI to existing job descriptionsAI mentioned in ad text
3. Funding and growth cycle of tech employersFewer or more tech openingsTech and developer share of ads
4. Local ad-writing normsAds list credentials, or describe outcomesWording shares
5. Which audience each job site servesA site shifts toward frontline or tech jobsEvery share above, and comparisons between countries

The last row matters most. When a site changes its audience, every share we count moves without any change in the labor market. Findings 4 and 5 are both about this row.

Finding 1 · How large is it?

Global employers added AI to every kind of job

Share of Hacker News "Who is hiring?" posts (tech employers, mostly US) that name AI tools or AI work, by job function.

What it shows. Marketing moved from 0.7% to 25.2%, recruiting from 0.3% to 14.5%. The largest relative jumps are outside engineering.

One level deeper. Dedicated AI roles also grew, but from a small base. In the same source, AI roles by title went from 0.83% of posts in 2019–21 to 7.49% in 2024–26, and from 4.2% in early 2024 to 13.6% in the third quarter of 2026.

AI roles by job title, Hacker News hiring threads, 2011–2026. Same source and same counting in every era.

The same pattern holds in a second, independent pool. Among 2026 global remote ads, with AI vendors removed, the share naming AI tools ranges from 19.6% (support and success) to 38.1% (engineering and data). AI roles by title are 5% of that pool after correcting for classifier error.

Global remote job ads first seen in 2026, AI vendors excluded. Bars show employer-clustered 95% ranges. The vertical line marks the share of ads that are AI roles by title.

Evidence AI named in 19.6% to 38.1% of ads across families → Observation titles show a fraction of that → Mechanism employers add a tool requirement to an existing job instead of opening a new one → Implication skill surveys and job-title counts understate AI demand.

Finding 2 · Where is it happening?

Indonesian ads did not move, on any of three boards

Share of Indonesian ads whose title or text mentions AI, archived from Kalibrr, Glints and loker.id. 4,764 ads before ChatGPT, 6,640 after.

What it shows. Pooled, the share went from 1.9% to 1.7%; AI roles by title from 0.36% to 0.35%. The pooled sample could detect a change of 0.31 points, so a move like the global one would have shown up many times over.

One level deeper. The gap holds when we compare like with like. With classifier error corrected, AI roles are 0.6% of Indonesian board ads against 5% of global remote ads, a gap of 8.1x (95% range 5.1x to 13.8x). Across four counting rules the gap stays between 2.8x and 6.9x.

Ratio of global to Indonesian AI-title share, log scale. Bars show 95% ranges (employer-clustered; the corrected row also resamples the label audit). The keyword classifier's precision was 60.5% on Indonesian titles and 71.3% on global ones, which is why the corrected row is the headline.

Evidence flat shares on three boards with enough power → Observation no lagged start visible by 2025 → Interpretation either adoption is absent, or Indonesian employers adopt tools without writing them into ads (see scenarios) → Implication an Indonesian ad that names AI stands out.

Finding 3 · Why is it happening?

Indonesian ads ask for credentials; global ads ask for outcomes

Share of ads using each kind of wording. Indonesia: three boards, 2023–25. Global: remote ads, 2026. Different years, so read the direction, not the exact gap.

What it shows. Ownership, stakeholder management and automation wording are four to nine times more common in global ads. Communication and problem solving are similar. Degree requirements run the other way: 48.9% of Indonesian ads against 10.3% of global ones.

One level deeper. AI requirements in global ads arrive inside outcome language ("own the workflow, use the tools"). An ad written as a list of credentials has no natural place for a tool requirement. That may explain part of finding 2 without any difference in actual AI use. SPECULATION

A reviewer judged 76.7% of sampled Indonesian wording matches correct, so treat small Indonesian shares as approximate.

Finding 4 · Who is affected?

The drop in Indonesian developer ads happened on Glints and loker.id

Software-engineering share of archived job titles, before ChatGPT and in 2023–25. The v1 draft reported this pooled as a 37% fall; split by board it looks different.

What it shows. Pooled, the share fell 4.14 points. Changes inside boards account for a fall of 4.98 points; the shift in board mix pushed the other way by +0.49 points. Kalibrr, the most professional-leaning board, shows no change (ratio 1.00×, range 0.73–1.31).

One level deeper. On Glints, the space developers left was filled by frontline work. Kalibrr shows nothing comparable.

Ad counts in the earlier then later era. Titles ranked by weighted log-odds with an employer-clustered interval.

Evidence decline concentrated in two boards, Kalibrr flat → Mechanism a board changing its audience (and possibly the 2022–23 tech-funding pullback) → Not supported AI substitution, since AI roles stayed under 1% → Implication developer supply on Kalibrr-type channels is as contested as before.

Finding 5 · Can countries be compared?

Compare countries on the same job site, or the ranking misleads

Pooled, India looks the most tech-heavy and Indonesia the least. That order mostly reflects where each sample came from: seven of ten countries are 82–98% company career pages, which list tech-company openings, while 1.6% of Indonesia's ads come from career pages. Hold the site fixed and the order flips.

What it shows. On Careerjet, Indonesian ads are 48.4% tech and US ads 16.2%, the reverse of the pooled order. Inside Indonesia, the site alone moves the answer from 2.8% on Glints to 48.4% on Careerjet. A cross-country "share of tech jobs" mostly measures which sites were read.

Tech share of ads in the 2026 cross-section, 61,419 postings from 28 sources, 14 Sep to 4 Oct 2026. Ranges and the full source mix per country are in the data file.

Finding 6 · What do employers ask for now?

AI skills and broad "engineer" roles are rising; single-tool skills and "developer" titles are fading

Same Hacker News "Who is hiring?" posts as Finding 1: tech employers, mostly US startups. Share of posts that name each skill, before ChatGPT and after.

What it shows. Large language models went from 0% to 10.4% of posts, AI agents from 0% to 7.2%. Asking the hire to own the result rose from 3.6% to 9.8%. Named programming tools fell: Java from 8.7% to 3.8%, Ruby from 10.5% to 4.7%. React is still one of the most requested skills, at 22.3%, but down from 30.9%.

One level deeper. The shift is from naming tools to describing people. Posts using generalist wording ("wear many hats", "full-stack", "end-to-end") rose from 30.5% to 40.3%; posts hiring for one named specialty, such as backend, iOS or DevOps engineer, fell from 30.2% to 25.3%. ANALYSIS

Job titles tell the same story. Rate per 1,000 hiring posts carrying each title, 2011 to Oct 2022 against 2023 to 2026.

Titles ranked by weighted log-odds with an employer-clustered interval; 1,000 posts is the unit. Titles are as employers wrote them, so "full-stack engineer" and "full stack developer" are counted apart.

What it shows. "Founding engineer", the first engineers at a startup, went from 2.0 to 21.2 per 1,000 posts. "AI engineer" barely existed before ChatGPT (0.1) and is now 4.2. The same job splits by word: "full-stack engineer" rose from 7.3 to 13.0 while "full stack developer" fell from 7.0 to 3.8. Front-end, web and Android "developer" titles all but disappeared.

Caveat: Hacker News hiring posts lean toward US startups, which hire founding engineers and adopt new tools early. Read these as the direction among tech employers, not the whole labor market. Skill matching is literal word matching on ad text; reviewed terms were spot-read.

Evidence AI skills and ownership wording up, single-tool requirements down → Mechanism AI tools let one person cover more of the stack, so employers hire for range and results → Implication a CV or an ad built on a list of tools reads as dated.

Contradictions

Things that are true at the same time

• AI demand is everywhere; AI jobs are rare. 19.6%–38.1% of global ads name AI tools, while 5% are AI roles.
• "AI" appears in most ads, rarely in titles. In 2026 the term shows up somewhere in 30.2% (Singapore) to 80% (India) of career-page ads in the nine other countries (mostly career pages), often in the employer's own pitch, while 2.6% to 7.9% of titles name an AI role. Treat the larger number as an upper bound on demand.
• Indonesian developer ads fell; Indonesian developer demand may not have. The pooled share fell 4.14 points, while Kalibrr held flat.
• Indonesian ads look junior-friendly, partly by language. Fewer Indonesian titles say "senior", but Indonesian titles rarely carry seniority words at all, so the seniority comparison is not reported here.

Second- and third-order effects

What follows from the findings

1. Ifthenand then
2. AI is a requirement inside jobs Finding 1Candidates are screened on tool fluency in roles that were never "AI jobs"People in marketing, recruiting and support who show tool use gain an edge, and training budgets follow
3. Indonesian ads stay silent on AI Finding 2Indonesian candidates meet a wording standard in global remote ads that local ads never trained them forLocal employers who do name AI tools draw a self-selected, more tool-fluent applicant pool
4. One site's audience shift drives national-looking trends Finding 4Any "Indonesian tech jobs fell" claim built on aggregators is fragileSite-level panels, not pooled shares, should anchor policy and media claims

Risks and alternative explanations

Red team: what could make us wrong

Scenarios, not predictions

Where Indonesian engineering ads could be in 2027

Starting point: 2.8% of Indonesian engineering ads named AI tools or AI work in 2023–25 (Kalibrr alone: 3%). Global engineering ads reached 23%. Each range below is arithmetic from the stated assumption. MODEL

Flat lines are era averages, not yearly values. Each cone runs from the 2023–25 Indonesian level to one scenario's 2027 range. Ranges, not forecasts.

Sensitivity. The headline gap is the least sensitive number in the report: four counting rules keep it between 2.8x and 6.9x. The scenario ranges are the most sensitive: changing the lag from 4 to 6 years moves scenario A by about five points.

Who wins, who loses

Distributional effects

Implications

What to do differently

What to watch next

Four numbers that would change our view