Summary: OMB Memorandum M-25-21 ("Accelerating Federal Use of AI through Innovation, Governance, and Public Trust") directs agencies to remove barriers to AI adoption and to develop an AI-capable workforce - not just AI practitioners, but everyday staff. If you’re a training officer, CAIO staffer, or team lead, the practical question is what "AI literacy for non-practitioners" actually looks like and how to deliver it this fiscal year.
What M-25-21 actually says about the workforce
Most of the memo’s coverage focused on governance: Chief AI Officers, use-case inventories, minimum practices for high-impact AI. But the workforce language is the part that lands on training calendars. Agencies are directed to invest in an AI-ready workforce - including AI literacy for employees who will use AI rather than build it, so they can work with these tools effectively, critically, and safely.
Pair that with the OneGov tool deals that put frontier AI assistants on federal desktops for next to nothing, and the situation at most agencies is: the tools arrived first, the skills are supposed to catch up, and "go watch a webinar" is the current plan of record.
What "AI literacy for non-practitioners" means in practice
A useful working definition - the one I train to - is that a non-practitioner is AI literate when they can do four things:
- Explain what generative AI is and is not - including why it produces confident wrong answers, and what that means for relying on it.
- Use the approved tools on their actual job - drafting, summarizing, rewriting for plain language, structuring analysis - not toy examples.
- Apply the guardrails - what data can and cannot go into a model, records implications, attribution, and when human review is mandatory.
- Spot opportunities - identify which parts of their own workflow are good AI candidates and which are not.
Notice what’s missing: prompt-engineering trivia, model architecture, vendor feature tours. Literacy is about judgment with the tools, not knowledge about them.
Why generic e-learning underdelivers here
Most agencies’ first move is a self-paced module. It checks the compliance box and changes almost nothing about how people work, for one reason: AI skills are motor skills. The judgment in step 3 and the opportunity-spotting in step 4 only develop when people use the tools on their own work products with someone experienced watching and correcting.
The pattern that works - in government and out - is short instructor-led sessions where participants bring real (sanitized) work: the memo they’re drafting, the comment file they’re summarizing, the report they produce monthly. Two hours of that beats twenty hours of videos.
What a team-level response looks like
You don’t need to wait for an agency-wide academy. A defensible team-level plan:
- Baseline (half a day): instructor-led AI fundamentals on government-shaped tasks, including the guardrails conversation with your agency’s actual policies in the room.
- Applied (two half-days, for teams ready to go further): build AI-assisted versions of the team’s three biggest workflows, stand up a shared prompt library the team owns, and leave with a 60-day adoption plan.
- Measure: hours saved per workflow, tracked by the team itself - which is exactly the evidence your CAIO’s office wants for the use-case inventory anyway.
Courses sized this way also fit under the micro-purchase threshold, which means a GPC card or SF-182 covers it without a contract action.
Questions to ask any AI training vendor
- Will the exercises use our work products and our approved tools, or your demo environment?
- How do you handle the guardrails module - generic "be careful," or our agency’s actual data and records policies?
- What documentation do we get for training files and the SF-182?
- Is the price published and fixed?
- What happens after the session - is there an adoption plan, or does everyone just go back to their inbox?
When this matters - and when it doesn’t
This applies if: your agency has AI tools deployed (or arriving) and your team’s training plan for them is currently a webinar link. Federal, state, and local all face the same gap; only the memo’s letterhead differs.
Skip this if: your agency has a real instructor-led AI literacy program with role-based tracks already running. Some do. Use it.