How AI work is taking shape.

Three findings

1. 929,173 qualifying job ads from 2,350 employers observed in 2026, including closed ads. 42,988 ads have a matching accepted AI review.
2. 2,331 ads in that AI-review subset include pay disclosed by the employer. All amounts are annual; conversions are labeled and salary estimates are excluded.
3. 1,747 job ads outside core technical occupations name coding agents in work or skill requirements.

Where AI work appears

2026 ads with a matching accepted AI review · 42988 job ads · 1239 employers

Occupation families are descriptive classifications, not identical jobs.

Entry points and seniority

2026 ads with a matching accepted AI review · 42988 job ads · 1239 employers

Unknown levels remain visible. Technical and management tracks are separate.

Observed hiring by country

2026 ads with a matching accepted AI review · 42988 job ads · 1239 employers

A job counts once in each explicitly listed eligible country and once globally. Country totals overlap and must not be added. Regional/worldwide eligibility is kept separate; countries are never invented from a broad region.

Observed hiring across employers

2026 ads with a matching accepted AI review · 42988 job ads · 1239 employers

Top employer domains in current reviewed AI postings. Business-model classification remains incomplete; tool mentions never determine employer type.

Annual advertised pay by occupation

Employer-disclosed pay in United States · 807 job ads · 97 employers

United States · USD per year · base salary. Separate lower/upper-bound medians; not employee earnings. All pay shown annually: monthly ×12, weekly ×52, hourly ×disclosed hours or assumed 40×52. 17 records use assumed hours.

Annual advertised pay by seniority

Employer-disclosed pay in United States · 807 job ads · 97 employers

United States · USD per year · base salary. Descriptive; occupation mix varies. Internships are shown in the adjacent panel when their separate pay population meets the sample rules. All pay shown annually: monthly ×12, weekly ×52, hourly ×disclosed hours or assumed 40×52. 17 records use assumed hours.

Internship pay: annual equivalents

Internships and apprenticeships in United States · 50 job ads · 23 employers

United States · USD per year · salary as advertised; base component not separately stated. Internships and apprenticeships; separate lower/upper salary-bound medians. This population stays separate from the seniority chart's pay basis. Annual equivalents compare pay rates over a full year; they are not earnings for a shorter placement. Monthly ×12, weekly ×52, hourly ×disclosed hours or assumed 40×52; 29 use assumed hours.

Distribution of advertised lower bounds

2026 ads with a matching accepted AI review · 807 job ads · 97 employers

United States · USD per year · base salary. Annual lower bounds only; equal-width bins of 30,000, lower-inclusive and upper-exclusive. No range midpoints. All pay shown annually: monthly ×12, weekly ×52, hourly ×disclosed hours or assumed 40×52. 17 records use assumed hours.

Annual advertised pay by city

Employer-disclosed pay in United States · 807 job ads · 97 employers

United States · USD per year · base salary. Cities above sample and employer-diversity gates. All pay shown annually: monthly ×12, weekly ×52, hourly ×disclosed hours or assumed 40×52. 17 records use assumed hours.

AI tools named in job ads

All qualifying job advertisements · 929173 job ads · 2350 employers

Each family counts once per job: ChatGPT (OpenAI) includes Codex and OpenAI/GPT; Claude (Anthropic) includes Claude Code. Company promotion, selling and restrictions are excluded.

AI and developer tools across professions

All qualifying job advertisements · 929173 job ads · 2350 employers

Required/preferred/workflow contexts only. Percentages use occupation denominators. Zero means no eligible mention; separate hybrid professions remain in downloadable data. VS Code and CLI alone do not prove AI use.

What a tool mention actually means

All qualifying job advertisements · 929173 job ads · 2350 employers

Each tool family counts once per job. A job may name more than one family.

Tools advertised together

All qualifying job advertisements · 929173 job ads · 2350 employers

Families named together in work or skill requirements. Each pair counts once per job; no model is inferred from an editor.

AI tools required outside AI-tagged jobs

Jobs not classified as AI · 886185 job ads · 2189 employers

Required skills only. Denominator: 886,185 job ads not classified as AI. These include unreviewed, stale and non-accepted reviews; they are not a verified non-AI population. Background text, selling and restrictions do not count.

AI tools inside and outside AI-tagged jobs

All qualifying job advertisements · 929173 job ads · 2350 employers

Required skills, preferred skills and advertised work, counted once per family per vacancy. Each row uses its own full job denominator, including ads with no tool mention. Outside the AI cohort does not mean verified non-AI. Descriptive rates; role and employer mixes differ. Use Explorer to compare the same profession, level and country.

The evidence funnel

2026 ads with a matching accepted AI review · 42988 job ads · 1239 employers

Each stage has its own population and denominator. Salary eligibility never depends on an estimate.

Main AI countries: average annual pay

Separately scoped national paid-wage and advertised-pay populations · Annual pay in each country display currency; means and medians labeled separately

Country-by-country reference values retain native display currencies, separate populations and reference periods. No all-sector multinational average or AI pay premium is computed.

Our job observations in a wider global context

Artificial Careers vacancies and distinct official populations · Separate job-ad counts, rates and official reference units

Our rates count distinct job advertisements requiring an AI tool, divided by all eligible ads for that country. Generic editors and command-line mentions alone do not count. Multi-country ads count once in each eligible country, once globally. Official indicators retain their own population, unit, source and reference year; they do not measure the same thing as our job-ad rates. Missing references are unavailable, not zero. No official statistic enters confirmed salary calculations.

AI tools beyond software teams

Business occupations in our catalogue; separate seven-country SME survey · Distinct job ads and a separate SME-survey percentage

Our profession rows count distinct job advertisements requesting AI tools in required skills, preferred skills or advertised work; the required-only column stays separate. Each profession uses its own full vacancy denominator. The separate OECD survey covers business functions among GenAI-using SMEs in seven countries in 2024, excluding France. Occupations and business functions are not one-to-one matches; do not subtract these percentages or interpret job mentions as company adoption.

Methods and limitations

• Vacancy, tool and salary populations retain the previous complete release. Official context was retrieved separately on 31 October; each reference keeps its original period, source, population and quality flags. These comparisons add no salaries and do not establish national AI adoption or an AI salary premium.
• Annual unique vacancy records observed in our catalog during the year, including active and closed advertisements. Observation dates do not establish original publication dates or a year-long hiring census. This is not a sum of monthly stocks or a count of hires. Current stored descriptions cannot establish historical monthly tool or pay curves.
• Observed job advertisements are a convenience sample, not national employment or actual organizational adoption.
• Role, employer and mention classifiers are conservative rules; independent human evaluation is required before public publication.
• AI review acceptance is source-linked classification, not an independently certified ground truth.
• Unknown location, seniority and employer facts remain unknown. No global salary pooling or missing-pay imputation.
• All salaries are annual. Monthly ×12; weekly ×52; hourly ×disclosed weekly hours ×52, otherwise explicitly assumed 40×52. Annualization preserves verified salary status; calculated totals and any assumed work hours are labeled.
• Historical salary and tool curves are withheld until comparable frozen evidence exists.
• Official AI adoption, occupational exposure, employment, internet and R&D references retain independent populations, methods, years and units. They are context, not matched comparisons or inputs to confirmed advertised pay. Job-ad tool rates and SME business-function rates must not be subtracted; neither establishes national tool adoption.
• This is a year-to-date local preview; not a completed annual publication.