I know I sound like a broken record, sorry, not sorry, but if you don't see the future of AI as a digital transformation project that changes the infrastructure of your organisation, rather than as the equivalent of launching a new website, you are missing the point.
The most useful sentence in Tony Holmes' recent comments to Federal News Network is the one underneath his AI literacy line. Agencies that get stuck between pilot and production, he says, almost always get stuck in the same place: they solved the technology problem but not the workforce one. The program manager ends up with a shiny new AI tool, does not trust it, and works around it. The contracting officer cannot evaluate a vendor's claims because nobody taught them what questions to ask.
If that sounds familiar, it should. It is the same pattern playing out in banks, hospitals, councils and consultancies right now. The tool lands. The pilot works. Then the wider organisation is asked to absorb something it has never been prepared for, and it does what humans always do with unfamiliar systems they do not understand. It routes around them.
Holmes, who leads solutions consulting at Pluralsight, made his case at the AI and Data Exchange 2026. The line worth pinning above your desk is this one: "you can't scale easily if you can't figure out what success looks like, and you can't measure what you haven't defined".
Now go and look at how your organisation currently measures AI literacy. If the answer is course completion rates and a feel-good NPS score after a one-hour webinar, you have defined attendance instead of success.
This is the gap most enterprise training programmes are sitting in. A single "AI 101" module gets pushed out to a workforce containing fifty different jobs, each with a different relationship to AI and different decisions they need to be capable of making. A claims handler needs to know when to override a model's recommendation. A director of operations needs to know how to interrogate a vendor's accuracy claims. A board member needs to know which questions to ask before signing off a procurement. Those are different skills. Pretending they are the same is how you end up with a workforce that has technically been "trained" and practically cannot use the thing.
The Labor Department's AI Literacy Framework, published in February, is one of the more sensible attempts to fix this. It treats literacy as something role-shaped rather than generic. Holmes called it "a real gift to agencies", and he is right, but the lesson generalises well beyond US federal work. Anyone designing AI capability programmes should be asking: what does competence look like for this role, in this context, on this decision? If you cannot answer that, you cannot teach it, and you certainly cannot measure whether you taught it.
There is a second point in his interview that I think gets less attention than it deserves. Holmes describes AI literacy as "a mission readiness posture" rather than a training budget item. That is a meaningful reframing. Treat it as L&D spend and it competes with every other training line, loses, and gets cut to a single webinar. Treat it as readiness, in the same category as cyber hygiene or safeguarding, and the conversation changes. It becomes a capability the organisation has to maintain rather than a course it has to deliver.
The agentic AI point at the end of his comments is the one that should worry boards most. Once autonomous systems start operating faster than governance boards can review them, clamping down produces shadow IT. People will use the tools anyway. They will just use them invisibly. The only durable defence against that is a workforce literate enough to make sensible decisions in real time, because the governance committee will not be in the room.
The practical question for anyone running people development right now is simple. Pull your AI training plan. Look at the success metric. If it is completion, you have not started yet. Role-mapped definitions of competence, written down before the curriculum is built, are the unglamorous bit of work that decides whether any of this actually lands.
One thing to try this week: pick three roles in your organisation and write a one-paragraph definition of what AI competence looks like for each. If the three paragraphs come out identical, your training programme has a problem your LMS will never surface.