Correction, 10 July 2026. The original version of this post claimed Indonesian IT job posts mention AI 28x less often than global remote roles. That was wrong. Two people left comments asking careful questions about the method, I went to check, and the number didn't survive. Here's the whole thing.
The number came from nowhere
What I published: 0.3% of Indonesian listings (3 of 1,039) carried an AI signal in the title, against 8.6% of global remote listings (216 of 2,517).
Neither figure exists in the article this post says it was drawn from. That article counts 1,039 local and 1,010 global — not 2,517 — and the words "28x", "8.6%" and "0.3%" appear nowhere in it. The multiplier got invented somewhere between the article and this post, and I never checked it against the source.
It gets worse. The May 2026 raw pull behind those counts was never saved, so nobody can re-derive them. Not you, not me.
And the CSV I linked at the bottom of the original post, as evidence? It contains 18 AI-titled roles out of 478. That's 3.8%, not 0.3%. Three of them are titled, plainly, AI Engineer. Another is Agentic AI Engineer. Anyone who downloaded the file I offered them would have seen my own data contradicting my own headline. Nobody did, which tells you something about how far a good number travels before anyone opens the spreadsheet.
What the data actually says
Same title scan, run properly over both sides:
| Pool | n | AI in title |
|---|---|---|
| Indonesian IT (JobStreet + Loker.id, Jun 2026) | 478 | 3.8% |
| Global remote, tech titles (Contra, WWR, RemoteOK, HN, Adzuna, The Muse) | 1,035 | 19.5% |
5.2x. Real, but a different animal from 28x.
Valentin asked whether I was measuring two countries or two job boards. Mostly the boards. Watch what happens if I don't hold role type constant, and compare Indonesian IT listings against the entire global pool — sales, support, ops and all: 3.8% against 6.2% of 3,318 listings. The gap falls to 1.6x.
One decision about what to compare moves the answer more than threefold. That was his point, and he was right to make it.
Where the gap actually lives
Luis asked where it concentrates. It turns out: not evenly, and not where I'd have guessed.
| Family | Indonesia | Global remote |
|---|---|---|
| Data / ML | 0% (n=11) | 23.7% (n=59) |
| Fullstack | 13.0% (n=23) | 11.4% (n=70) |
| DevOps / Cloud | 0% (n=18) | 2.1% (n=95) |
| Senior / lead titles | 2.6% (n=151) | 17.4% (n=391) |
Inside ordinary engineering titles the gap mostly isn't there. Fullstack actually flips — Indonesian fullstack roles put AI in the title slightly more often than global ones do. DevOps is a rounding error on both sides.
What Indonesia is missing is dedicated data/ML titles existing at all, and AI language in senior titles: 2.6% against 17.4%.
Luis called this before I did. He said the story was signal visibility — that employers use titles to position against competitors, not only to describe work. That is precisely where the gap sits: in the roles companies use to posture, not in the roles that describe the daily job. An Indonesian fullstack dev and a global one are advertised as doing much the same work.
Several of those Indonesian cells are under n=25. Directional, not settled.
What's still wrong with this
- I still haven't held the channel constant. A remote-first aggregator self-selects for AI-forward startups in a way JobStreet doesn't. That's Valentin's real question and I can't answer it yet — I'd need Indonesian remote-first listings from a comparable board.
- The snapshots don't line up. Indonesian side is June; global side is today.
- Title-only. A team can be neck-deep in AI and never say so in a title. This measures advertising, not adoption. The original post admitted that, then ran a headline that ignored it.
- No random sampling, no confidence intervals.
The Indonesian figures come from the public CSV below — go check my arithmetic. The global figures come from a snapshot of my own corpus. That title-level export is up: global + Indonesia title-level export, ledgered.
The part worth keeping
A number that flatters the thing you already wanted to say is the one to check twice. 28x made a much better headline than 5x. That is exactly why it sailed through.
Thanks to Valentin Monteiro and Luis. The correction is theirs.
Free datasets (CSV, no signup): Indonesian IT — replicated · Global remote — preview
Underlying six findings: Indonesia IT vs Global Remote
Top comments (7)
Both points land. The stub-vs-full-text asymmetry is now a permanent guard in the pipeline — same shape as the dead dataset you're describing: one side is search-card fragments, the other full JDs, so a regex ends up reading population size where it should read population rate. Built a check that fails the build on exactly that mismatch (median-length ratio + empty-row fraction) after finding it live in a second dataset.
Your sample-size point checks out too. n=28 at ~4% base rate gives an expected count around 1 — the CI at that count swallows any real effect. Ran the numbers: a ±3pt margin at 4% needs on the order of 150–200+ clean rows, not 28. So the remote-first cut isn't close to ready; I'll wait for real N before publishing anything from it.
Really interesting dataset — the 28x gap is striking, even accounting for title-level noise and sampling limitations.
What stands out is less “AI adoption” itself and more signal visibility: global remote roles are explicitly marketing AI capability in the job title, while Indonesian listings seem to keep it implicit or absent. That alone creates a perception gap in the market, even if underlying usage is closer than it appears.
I also think your caveat is important — title-only classification will always undercount real AI usage — but as a hiring signal proxy, this is still very meaningful. Employers don’t just use titles to describe work, they use them to position against competitors.
Would be interesting to extend this by segmenting by role type (backend, data, frontend) or seniority to see where the gap is actually concentrated.
Your segmentation suggestion is what broke this open, so it gets a full answer — thank you.
First the correction: the 28x was wrong. It appears nowhere in the source analysis this post was repurposed from, and the Indonesian CSV I linked as evidence shows 18 AI-titled roles out of 478 — 3.8%, not the 0.3% I claimed. Three are titled literally "AI Engineer". The post now carries a correction at the top. Corrected gap: ~5.2x (3.8% Indonesian IT vs 19.5% of global remote tech titles).
Now your actual question — where the gap concentrates. It is emphatically not uniform:
Inside conventional engineering titles the gap mostly disappears — and for fullstack it inverts: Indonesian fullstack roles put AI in the title slightly more often than global ones. What's missing from the Indonesian pool is (a) dedicated data/ML titles existing at all, and (b) AI language in senior titles: 17.4% vs 2.6%.
That is your hypothesis, not mine. You argued the story was signal visibility and competitive positioning rather than adoption. Segmented, that's exactly what shows up: the gap lives where employers are positioning — senior and specialist titles — not where they're describing day-to-day work. An Indonesian fullstack dev and a global one are advertised as doing roughly the same job.
Two caveats. Several Indonesian cells are n < 25, so treat those as directional. And I still can't hold the channel constant — a remote-first global aggregator self-selects for AI-forward startups in a way JobStreet does not. That's Valentin's confound, and it's unsolved.
Your other point deserves a flag. You wrote that title-only classification will always undercount real AI usage. I had built a second dataset to fix precisely that by classifying full job descriptions — and checking your question is how I found it's broken. The Indonesian descriptions in it have a median length of 18 characters (they're search-result stubs); the global side has full descriptions. A regex classifier reading long text on one side and stubs on the other invents a gap out of nothing. It never got published. It won't.
Interesting analysis. The gap between local IT opportunities and global remote roles is something many developers are experiencing.
I think AI is changing this landscape quickly. Developers who only compete on coding skills may face more pressure, but those who can combine software engineering + AI tools + business problem solving will have much stronger opportunities.
From what I’ve seen, many companies are no longer just looking for someone who can write code — they want engineers who can use AI to build faster, automate workflows, improve products, and communicate effectively with global teams.
For developers in Indonesia and other emerging markets, remote opportunities become much more accessible when they build:
The future advantage will belong to developers who treat AI as a productivity multiplier, not a replacement.
Great data and discussion. Looking forward to seeing more insights about how AI will reshape remote hiring globally.
The 28x is striking, but I'd wonder how much of it is the two pools rather than the two countries. A global remote board self-selects for startups and English-first, AI-forward employers, while JobStreet Indonesia pulls a much broader enterprise/SME mix, so part of what you're measuring is platform composition. A cleaner cut might be Indonesian remote-first listings against the same global remote pool, or the same role families on both sides. Do your datasets let you hold the channel roughly constant, or is the platform kind of baked into the geography here?
You were right, and chasing your question turned up something worse than composition — so thank you.
Short version: I checked, and the 28x itself was wrong. It appears nowhere in the source analysis this post was repurposed from, and the Indonesian CSV linked at the bottom of my own post shows 18 AI-titled roles out of 478 (3.8%), not 3 of 1,039 (0.3%). Three of them are titled literally "AI Engineer". I've rewritten the post with a correction at the top.
On your actual question — yes, the platform is substantially baked into the geography, and I can now put a number on it. Same title scan, both sides:
So just choosing whether to hold role type constant moves the answer by more than 3x. Composition was doing most of the work in the original number, exactly as you suspected.
What I still can't do is hold the channel constant. A remote-first aggregator self-selects for startups and English-first employers; JobStreet doesn't. Your suggested cut — Indonesian remote-first listings against the same global remote pool — is the right one, and I don't have the data for it yet. I have a scrape with ~28 remote-flagged Indonesian rows, which at a ~4% base rate is far too underpowered to say anything. Collecting a proper Indonesian remote-first sample is the next thing I'll do, and I'll publish it whichever way it comes out.
One more thing your question shook loose: a second dataset I'd built to fix the "title-only" limitation is also broken — the Indonesian job descriptions in it have a median length of 18 characters (search-result stubs), while the global side has full descriptions. A regex classifier reading long text on one side and stubs on the other will manufacture a gap out of nothing. That one never got published, and now it won't.
Respect for chasing it down and correcting at the top rather than quietly editing the number. The stub-vs-full-text catch is the one I'd frame as the real lesson: any classifier reading 18 chars on one side and full descriptions on the other is measuring input length, not the thing you think you're measuring, and that asymmetry kills more cross-market comparisons than composition ever does. For the remote-first cut, the base rate is your real constraint, not the sample size: at ~4% you need a few hundred clean rows a side before the ratio means anything, so 28 was never going to talk. Worth the wait to collect it properly.