AI capability is what remains when the novelty wears off.

A working definition

AI capability is the demonstrated ability to produce professional work through a redesigned AI-native workflow — one that integrates human expertise, AI, tools, explicit standards, verification and evidence — repeatedly, and to a stated quality. Operating that workflow is what makes a professional AI-augmented.

Every word in that definition is doing work.

• Demonstrated, not self-reported. There is a record.
• Redesigned workflow, not the old process with AI bolted onto one step.
• Integrates — human judgment and AI contribution are allocated deliberately, not by default.
• Explicit standards: the definition of good is written down before the work starts.
• Repeatedly — once is an anecdote. A system produces the next one too.
• To a stated quality — measured against something, not against a feeling.

What that looks like in practice

• Choosing which parts of the work AI should touch, which it should not, and being able to say why.
• Checking an output against a written professional standard before it leaves you, not against a feeling.
• Running a repeatable workflow in which who owns each decision is visible, not assumed.

Why measurable

Not because measurement is inherently virtuous, but because the alternative has failed. Organisations have spent three years running AI pilots that produce enthusiasm, then dissipate. Individuals have spent three years accumulating impressive outputs and no transferable skill. In both cases the missing element is the same: nothing was ever baselined, so nothing could be improved, defended, or built on.

If you cannot say where you started, you cannot say whether you moved.