AI literacy is often translated into a training: a module, a certificate, a box ticked. That is not incorrect, but it is not what the word calls for either. Literacy means that someone understands what a system does and does not do, where the limits of an answer lie, and when an outcome needs to be checked before it ends up somewhere else. That is a different skill from being able to operate a chat window.
An employee who knows how to formulate a prompt is not automatically literate in the sense that matters here. AI literacy is about recognizing risk: when does an answer contain an error that is not noticed, when is company data shared with a system that uses that data elsewhere, when does an AI outcome form the basis for a decision that actually deserves review. Failing to teach that means teaching employees to operate a tool without teaching them when they should not trust it.
Literacy is only meaningful if it connects to what happens in practice. If employees are already using a tool that no one has approved, then a generic training on AI principles is worth little as long as no one knows which tools these are and what they are used for. Literacy without insight into usage is a program that misses the practice entirely.
That is also where the easiest way to learn something lies: asking. Not asking to hold someone accountable, but asking to understand what is going on. An employee who uses a tool to work faster rarely volunteers that information if the question feels like a check that could lead to a reckoning. Literacy therefore does not begin with giving instructions, but with creating a climate in which reporting carries no risk.
A training can explain what a language model is, where bias comes from, and when an output needs to be checked. A training cannot guarantee that this knowledge is applied at the moment it matters: under time pressure, ahead of a deadline, for a task that has already gone well a hundred times. Nor can a one-off training keep up with what changes. Systems get adjusted, new tools become available, usage shifts. That is why literacy does not belong to a single moment, but to a recurring rhythm — just as reassessing risk classifications is not a one-off exercise but a recurring process.
Literacy among employees does not relieve an organization of the question of who oversees what. If an employee incorporates an AI outcome into a decision, and that decision turns out to be incorrect, then the question of who is responsible when an AI application makes a mistake is not answered with "the employee should have known." Responsibility lies at multiple levels: with whoever deployed the system, with whoever organized the oversight, with whoever set the frameworks. Literacy among individuals is part of that chain, not a replacement for it. What human oversight in practice means is connected to the question of whether the person exercising that oversight also sufficiently understands what they are overseeing — without that understanding, oversight is a formality.
Many organizations try to link AI literacy to an existing compliance or risk process. That seems logical, but in practice a second layer alongside an existing process is often not used: employees follow the process they already know and ignore the new one, even when it fits the practice better. Why a second process alongside the existing one gets ignored is a question that goes beyond training: it concerns where literacy is documented, tested, and used in the work itself.
We cannot say how much training is sufficient, or what percentage of employees actually act differently after a program. That differs per organization, per role, per risk level of the tasks someone carries out. Someone who drafts client contracts with AI support on a daily basis needs a different level of literacy than someone who uses AI incidentally for a summary. A single answer to "how much training is enough" does not exist, and we therefore do not give that answer.
Literacy without insight into the risks of individual usage situations remains abstract. An employee who enters company data into a free chat window runs a different risk than an employee who uses an internal, shielded system, and what you do about company data in a free chat window is a question that requires a different answer depending on the situation. Literacy is therefore not a single program for the whole organization, but a match to what people are actually working with.
That match starts with knowing which tasks are already supported by AI and to what extent. The work scan from FTE TO AI calculates, per task, what portion of the work can be taken over by AI, giving a concrete picture of where literacy is needed most: not in a general sense, but at the place where the work is actually changing.
Vraag maar. Governance begint bij weten wat er draait — ook wat niemand heeft goedgekeurd.
Answers come from this site’s knowledge base. Not tailored advice, and not a scan of your company.