Quarterly · all markets · 24,944 live specifications · live basis
The distance between AI adoption and role design, measured every quarter.
The Verellax Index measures the distance between AI adoption and role design across 24,944 live job specifications from 217 employers across all markets, scored on four tests and published each quarter.
Published on a date we state in advance, and never edited after.
What does Edition one find?
Edition one · Q3 2026 · preview read as at 14 September 2026
Edition one publishes 1 OctoberCompanies adopted AI faster than they wrote it into the job. Of 24,944 live specifications from 217 employers, 5,674 are AI-central, roles where AI is core or supporting the work as written, and adoption climbs with seniority: 6.0% of entry-level roles are AI-central against 24.1% at director, as at 14 September 2026.
Of those 5,674 AI-central roles, 57.3% name nobody to overrule the model, own its output or hold a decision. The seat that answers for the model is the one hired least.
A designed minority exists in every cohort. The deficit is a choice rather than a property of the technology, and the organisations that own it are the ones writing the specifications, not the market they hire in.
60.3%of postings that mention AI hire a user of the model, from 51.3% in Q4 2025. Roles hired to govern it stayed between 2.5% and 4.3%.all employers · Q4 2025 to Q3 2026 · within-quarter shares · posting-date basis
The designed minority
42.3%of AI-central roles at the major Australian banks, insurers and super funds name no decision architecture. At the agency networks and independents, 71.1%. Across all markets, 11.0% of 12,547 roles that mention AI name who can overrule the model.AU banks, insurers and super funds n 123 · agencies n 432 · live basis
Entry level
6.0%of entry-level live roles are AI-central. At director level, 24.1% are: 4.0×, same markets, same month. The ladder starts above the graduate.entry n 2,185 · director n 1,865 · all markets · live basis
Adoption reached the executive layer and stopped above the graduate.
The Index reads every AI-central specification for decision architecture: who works with the model, who can overrule it, who owns its output, which decisions stay human. This is the share that names none of the four.
Read the four testsof AI-central live roles name nobody to overrule the model, own its output or hold a decision.
all markets · 5,674 AI-central live specs · live basis · as at 14 September 2026How is a specification scored?
The Index reads four properties of every specification and scores each one on four tests. The four properties: how central AI is to the role (core, supporting, peripheral, absent); how the work divides across the four layers; whether the role is a new title or a backfill; and whether it names decision architecture, meaning the four tests.
217 employers · 24,944 live specifications · live basis · as at 14 September 2026
Read the methodFour grades, read from the specification as written: core, where the role exists to work with the model; supporting, where the model is part of the work; peripheral, where it is mentioned but incidental; absent. Core and supporting together make a role AI-central. Of 24,944 live specifications, 5,674 are AI-central.
Every skill in the specification is classified into one of four layers: execution, coordination, judgement, or AI-adjacent. The layer shares describe the composition of the role, not a trend. Mapping coverage is stated per layer and is not yet flat across the corpus.
A new title is a role written for the first time; a backfill re-posts an existing one. The distinction separates roles redesigned around AI from AI folded into an old title, which is the redesign gap. Read on posting-date basis against the employer's own record.
A specification has decision architecture when it names who holds authority when the model is wrong. Four tests, each pass or fail: the person works with the model; someone can overrule it; someone owns its output; at least one decision is written as the person's to hold. Of 5,674 AI-central live roles, 57.3% pass none.
Which markets and sectors does it cover?
ten markets, led by United States, Australia, United Kingdom, Germany and Ireland, across six sectors: financial services, technology, AI-native, advertising and agencies, consumer products and professional services, on 24,944 live specifications from 217 employers as at 14 September 2026. The measure by market and by sector publishes on 1 October 2026; until then the chart shows where the record comes from.
all markets · six sectors · 217 employers · live basis · as at 14 September 2026
How often is it published, and what changed?
Figures re-render weekly from that week's pack. The Index re-publishes each quarter on a stated date. Edition pages never change; corrections append as dated notes.
Preview · re-reads each Monday from the weekly pack · publishes · edition two
Edition one publishes 1 OctoberHow do I cite the Index?
Cite the edition and the as-at date. Published under CC BY 4.0. Corrections append to the edition page as dated notes.
Licence CC BY 4.0 · pending confirmation · corrections append as dated notes
Send a correctionVerellax Index, Edition one, Q3 2026. Verellax workforce intelligence, 24,944 live specifications, 217 employers, all markets, as at 14 September 2026. verellax.com/index/edition-one