Skill adoption in tech hiring, measured over nine years

Every skill that became a mainstream hiring requirement in nine years of Swedish tech job ads showed measurable posting momentum first: 7 to 28 months before it crossed the mainstream bar, 15 months at the median. We measured this on 413,133 Data/IT job ads published between January 2016 and December 2024, scored against 28 skills whose outcomes are now known with hindsight. The detection rule only ever saw data available at each simulated month. This issue reports the crossings, what the same rule got wrong, and the one thing the data cannot say.

Four skills crossed into the mainstream. All four were flagged early.

The mainstream bar is a 2% share of rolling-quarter postings, sustained for six months — calibrated to this corpus, where incumbents such as Java hold 8–15%. The momentum rule: a trailing-quarter share of at least 0.5%, at least 1.5× the skill's own 12-month baseline, on at least 12 ads.

Recall was 1.00: no skill crossed the bar without being flagged first. The rule stayed silent on hype that collapsed (blockchain peaked at 0.07% of postings and fell to zero) and on the long declines of jQuery, Perl and PHP.

What the rule cannot do

The same rule is not a forecast. Of the flags it raised that time has since judged, 29% preceded a crossing. It also fired on the 2023 recovery of React (mainstream since 2016), on durable skills that plateaued below the bar (Kotlin, Terraform, Ansible each settled at 1.3–1.7%), and on brief rebounds of declining ones (Hadoop in 2019, Scala in 2020). Growth against a 12-month baseline says a skill is moving. It does not say where the movement stops.

So this bulletin publishes momentum — the share now, the multiple over the baseline, and the sample behind both — and never a prediction of what becomes mainstream. A rule with these error rates supports the first claim and would be dishonest making the second.

The capital moves first

For the skills that rose within the measured window, Stockholm's posting share crossed adoption thresholds before the rest of Sweden did. (Skills already mainstream nationwide at the window's start, such as Python, leave no lead to measure.)

A lead measured in months to years means metro-level series carry information a national aggregate blurs — which is why our market dossiers track metros separately from their countries.

One external check

Sweden is one mid-size market, so we checked its cycle against an official US series. Quarterly year-over-year growth of Swedish Data/IT posting volume correlates with US BLS JOLTS Information-sector job openings at Pearson r = 0.77 over the 32 shared quarters. Both cycles peaked in 2021.

Measurement notes

• Skill counts are freetext concept matches over ad text: a proxy that is consistent over time, not extraction-grade attribution. The bias is constant, which trend detection tolerates and level claims must not quote unqualified.
• The source corpus has two collection-regime breaks (a downward level shift in 2020, a staged upward capture change in 2025), so scoring stops at December 2024. Run naively across the 2025 break, the engine flags the measurement artifact — harder than it flags any real trend. Our own collection therefore quarantines trend history across any coverage change and says so on the page.
• Thresholds calibrated on Sweden transfer to other markets as starting points, not constants. Each market's bars are re-derived from its own share distribution and volumes.

What we collect now

We collect tech job postings weekly for the United States and Australia, under the methodology above with a fixed collection cadence. Quarter-grain figures for each market publish once it accumulates a full stable quarter; the dossiers state their own measurement windows.

Data and citation

The monthly counts behind this study are published under CC BY 4.0 on the open data page. Swedish ad data derives from the JobTech (Arbetsförmedlingen) historical ads API; US openings from the BLS JOLTS series (public domain). Cite as: CodePrep, Global Tech Hiring Bulletin No. 1, July 2026.