Skip to content
Verellax

Three layers of human work, and one that operates the machine.

The taxonomy as applied to 24,944 live specifications as at 14 September 2026: 830 canonical skills, each in one of four layers.

Live basis · W38 · as at 14 September 2026 · 24,944 live specifications · 830 canonical skills
Execution24,940100.0% of 24,944 live · 512 canonical skills · the work a model absorbs first: producing, processing and completing defined tasks.
Coordination24,31497.5% of 24,944 live · 104 canonical skills · the work of moving it between people.
Judgement24,14196.8% of 24,944 live · 161 canonical skills · the work of deciding.
AI-adjacent10,79243.3% of 24,944 live · 53 canonical skills · the work of operating the model itself.
The skills-side mirror

Employers ask for the use of the model five times more often than they ask for the governance of it: ai tool use in 8,723 live specifications, ai governance in 1,596.

01

What is execution

The work a model absorbs first. Technical: building, coding, configuring. Functional: the defined tasks of a role, done. Domain: the knowledge the tasks require. The largest layer by canonical skills, 512 of 830, and the one where AI substitution is most direct.

Functional20,98384.1% of live · The defined tasks of a role, done.
Domain20,57782.5% of live · The knowledge the tasks require.
Technical16,78867.3% of live · Building, coding, configuring.
02

What is coordination

The work of moving work between people. Internal orchestration: inside the organisation. External orchestration: clients, partners, regulators. Enablement: making others able to do the work. Strategic alignment: connecting work to intent.

Internal orchestration22,71791.1% of live · Moving work between people inside the organisation.
External orchestration11,95247.9% of live · Clients, partners, regulators.
Enablement9,50338.1% of live · Making others able to do the work.
Strategic alignment4,80619.3% of live · Connecting work to intent.
03

What is judgement

The work of deciding. Operational judgement: decisions inside the task. Strategic judgement: decisions about which tasks. Evaluative judgement: deciding whether the output is right, the sub-category a model most needs a human for. Taste: the one a specification almost never asks for by name.

Operational judgement20,34381.6% of live · Decisions inside the task.
Strategic judgement11,74547.1% of live · Decisions about which tasks.
Evaluative judgement11,00844.1% of live · Deciding whether the output is right. The sub-category a model most needs a human for.
Taste9213.7% of live · The skill a specification almost never asks for by name.
04

What is AI-adjacent

The work of operating the model. Tool use: using it. Tool building: making it. Governance: deciding who is accountable for it. The governance sub-category is the skills-side mirror of decision architecture.

AI tool use8,72335.0% of live · Using the model.
AI tool building2,0428.2% of live · Making the model, or the systems around it.
AI governance1,5966.4% of live · Deciding who is accountable for the model. The skills-side mirror of decision architecture.
05

How the layers relate to the mark

The mark carries the four layers as four shapes on one frame: execution the ring at 3, coordination the ring at 7, judgement the crossing at 9, and AI-adjacent the six leaves that meet at the record. Alone, each shape is the layer's glyph; together they are the mark. The map on the skills hub is the same four shapes, counted by the data.

The map, on the hub
ExecutionThe amber ring at 3. Alone: one circle.
CoordinationThe teal ring at 7. Alone: two circles in line.
JudgementThe crossing at 9, teal with a signal outline. Alone: the crossing.
AI-adjacentThe six coral leaves meeting at the record. Alone: six leaves.
Alone, each shape is the layer's glyph; together they are the mark. Where shapes cross the upper wins: leaves over crossing, rings over both.
06

How this is measured

Live basis. Layer shares on the market pages are the average across specifications of the share of each role's work in that layer; the counts here are specifications carrying at least one skill in the sub-category.

Full method at /method
Source: Verellax workforce intelligence · 24,944 live specifications, 24,941 with a mapped skill · 263,750 skill observations · 830 canonical skills · live basis · as at 14 September 2026
Figures re-render each Monday from that week's pack.