Why does AI literacy matter for organisations?

The Gap the Measurements Keep Finding

Two findings sit side by side in the research, and they appear to contradict each other.

At the level of individuals and tasks, AI gains are real and well documented: drafting, research, analysis, and coding all show measurable acceleration in controlled studies, with the largest gains often going to less experienced workers. At the level of the firm, the picture inverts. The most comprehensive executive survey to date, published by the National Bureau of Economic Research in February 2026, covered roughly six thousand senior executives across four countries: more than 80% of firms reported no measurable productivity gain from AI over the previous three years. Separate work comparing what executives report against what revenue and employment data imply found the reported gains shrink on contact with the accounts.

The standard explanation is the productivity J-curve: general-purpose technologies pay off slowly, because the technology is the cheap part and the complementary investment — redesigned workflows, restructured data, new skills — is the expensive part that takes years. That explanation is fair, and it fits the history of electricity and computing well.

But "complementary investment" is doing a lot of unexamined work in that sentence. Workflow redesign and data infrastructure are the two everyone names. There is a third, and it is the one this page is about: most organisations are systematically discarding the most valuable thing their AI-literate people produce.

What Literate Work Actually Produces

Watch someone genuinely literate work with an AI system on a consequential task and you will see two outputs, not one.

The first is the artefact: the report, the analysis, the email, the code. It is visible, it gets filed, and it is what productivity measurement counts.

The second is everything that shaped the artefact: the context they supplied that the system lacked, the corrections they made when the output was plausible but wrong, the options they rejected and why, the judgement about what could be delegated and what couldn't. This output is the literacy — it is the judgement being exercised. It is also, in most organisations, invisible. It lives in the conversation, and the conversation is treated as exhaust.

Of the two, the second is worth more. The artefact solves today's task. The judgement that produced it is reusable: it is what the next person doing a similar task would need, and it is precisely the knowledge that was in nobody's documentation to begin with.

The Evaporation Problem

Here is the mechanism, made concrete.

An operations lead spends forty minutes working with an assistant on a difficult supplier email. Most of those minutes are not writing; they are teaching. We never promise refunds before inspection has confirmed the fault. This supplier is on a quality watch since March, so the tone needs to be firmer than usual. Don't reference the credit note until legal has cleared it. The final email takes two minutes to approve. It is sent, the task is done, and the forty minutes of organisational truth that produced it now exists in one place: a chat history that nobody, including its author, will ever open again.

Next week a colleague handles a similar dispute with the same supplier. The assistant they open knows none of it. The forty minutes is paid again — or worse, isn't, and the email goes out promising a refund before inspection.

This is the evaporation problem: each session begins from nothing and ends by discarding what it learned. The default tooling makes this structural, not accidental. Sessions are not built for organisational reuse. Where tools do remember, they mostly remember per person and per tool: the assistant learns its user's preferences, which helps that user and nobody else. What an individual gains, the organisation does not inherit.

Scale the example honestly and the firm-level survey results stop looking paradoxical. Individual time savings can be real on every single task while the organisation accumulates nothing, because the saving is consumed in the moment and the learning is destroyed at session end. Evaporation is not the whole explanation for the measurement gap — workflow and data investments are real constraints too — but it is a missing layer in the usual account, the part of the complementary investment that almost nobody has named, and the only part that is cheap to fix relative to re-architecting workflows.