AI Literacy
What is AI literacy?
The term is doing a lot of undefined work right now. It appears in board papers, L&D budgets, job descriptions, and European legislation, and in most of those places it quietly means the same thing: tool training. Show people the approved assistant, teach them to write a decent prompt, tick the box.
That translation is the problem this page exists to name. The people most likely to misuse AI in consequential work are not the ones who can't operate the tools. They are the ones who operate them confidently and can't tell when the output is wrong, what the system never saw, or which parts of the task should not have been delegated in the first place. Tool proficiency without judgement doesn't reduce that risk. It scales it, because fluent output gets trusted faster than clumsy output does.
AI literacy, defined honestly, is the thing tool training skips.
A Definition with Honest Scope
AI literacy is the ability to work with AI systems deliberately: understanding what a given system can and cannot do, deciding what to delegate to it and what to keep, communicating the task and its context well enough for the system to act on, and evaluating what comes back before it carries your name.
Each clause is there for a reason. Capability and limitation, because the systems are uneven in ways that don't announce themselves — fluent prose is not evidence of correct reasoning, and confidence of tone is not confidence of fact. Delegation, because the highest-stakes literacy decision happens before any prompt is written: whether this task, with these consequences, should go to a system at all. Communication, because most disappointing output is not a model failure but a context failure — the system was never told what the person assumed it knew. And evaluation, because the output is a draft of a judgement, and the judgement remains yours.
One useful articulation is the 4D AI Fluency Framework (Delegation, Description, Discernment, Diligence) by Prof. Rick Dakan and Prof. Joseph Feller, developed in partnership with Anthropic and freely available through Anthropic Academy. It is model-agnostic, which is the right property: literacy that only works on one vendor's tools was tool training all along.
What it is not
It is not prompt technique. Prompting skill is the most visible and most perishable layer — closely tied to how current systems behave, and partially obsoleted by each capability jump. The durable layer underneath is knowing what the system needs in order to act well, which survives every interface change.
It is not a course completed once. The systems change quarterly; a literacy programme built around one tool's behaviour in January describes a different tool by June. Training built around judgement — what to delegate, what to verify, what context was missing — ages far more slowly than training built around features.
And it is not certification against a standard, for an awkward reason worth knowing: nobody has managed to establish a widely accepted one. The EU's AI Act made AI literacy a legal obligation for providers and deployers from February 2025, originally requiring organisations to ensure a sufficient level of it among staff. The amendment provisionally agreed in May 2026 softens that to taking measures that support its development, in part because no one could say, across the range of organisations the law covers, what a sufficient level is. Whatever the final wording, the direction of both versions is the same: organisations are expected to develop this capability deliberately. The honest reading of the legislative wobble is that literacy behaves like a practice, not a threshold.
Engramic's approach
How Engramic Approaches it
There are multiple ways to support AI literacy in practice. This is one approach.
Engramic doesn't teach literacy, and a platform shouldn't claim to. What it does is give literate practice a structural home. The questions a literate person asks — what does this agent need to know, what context is it missing, what shouldn't it be trusted with yet — are, in Engramic, the questions through which an organisation's knowledge gets authored: answered deliberately, recorded, and available to the next agent and the next person rather than re-asked from scratch each session. The discernment stays human. What changes is that exercising it leaves something behind.