AI literacy creates AI capability

The argument

AI literacy is useful. It lets more people recognize opportunities, use tools safely and participate in redesigning work.

But an organization is not capable because many employees can prompt a model.

Capability means the company can repeatedly turn AI into a reliable outcome. That requires access to the right data, integration with real workflows, evaluation, ownership, budget, decision rights, security, escalation paths and a way to improve the system after it fails. Training can help with several of those. It does not create them automatically.

McKinsey’s 2026 work on scaling AI makes this distinction directly. It argues that the harder barriers are increasingly workflow redesign, operating models, leadership and culture, not simply access to the technology. Its work on agentic AI foundations adds data architecture and data quality to that list.

Literacy changes what people can try. Capability changes what the organization can repeatedly deliver.

Why people believe it

Training is visible and scalable. Leaders can count participants, completion rates and active users. It is also genuinely valuable, especially early in adoption.

The problem is using a learning metric as a proxy for an operating capability.

What the evidence says

Organizations that create more value from AI make structural changes around work, platforms, data and decision-making. The technology becomes embedded in a repeatable system rather than remaining dependent on a few unusually motivated people.

A useful test is transferability: can a second team reproduce the outcome without the original expert standing beside them?

The better question

Do not ask only “How many people have been trained?”

Ask: What can the organization now do reliably that it could not do before, and can another team reproduce it under normal operating conditions?

If the answer depends on one champion, one prompt library or one workshop, you have literacy. Capability comes later.