From AI Literacy to AI Agency

Executive Summary

AI literacy is becoming a national workforce priority. Employers, universities, workforce boards, and training providers are racing to help people understand and use generative AI. But a quieter, more consequential problem is emerging as a critical risk: access to AI does not automatically create agency with AI.

I’m watching people leave AI workshops inspired but not transformed. They learn a tool, complete a demo, copy a prompt, and feel a temporary boost of confidence. Then the training ends, the template disappears, the context changes, and the new capability evaporates. They struggle to transfer what they learned into new workflows, to evaluate AI outputs with discernment, or to decide what to delegate, what to co-create, and what to keep fully human.

This white paper argues that the future of work requires something different, a new approach. A deliberate shift from AI literacy to AI agency. Literacy teaches people what AI is and how tools work. Fluency helps them apply AI to real tasks. AI agency is the self-directed capacity to use AI with discernment, autonomy, judgment, creativity, and contextual intelligence across changing situations.

AI adoption goals should not just be for people to use AI. The goal is for people to become more capable because of how they use AI. This paper introduces Everyday AI Agency as a premier practitioner research agenda for studying how people build transferable AI fluency, preserve human judgment, reduce cognitive load, and expand personal capacity in an AI-enabled economy.

1. The Moment: AI Readiness Is Now a Workforce Priority

The national conversation has shifted from AI access to AI readiness, and federal institutions are now codifying what readiness means.

The U.S. Department of Labor released its AI Literacy Framework on February 13, 2026, to guide workers, employers, training providers, faculty, and workforce stakeholders while allowing adaptation across industries and contexts (U.S. Department of Labor). The framework states that literacy alone is insufficient: “In most workplaces, indicating a need for ‘AI literacy’ is not enough on its own; employers and other stakeholders may need to define the specific AI skills and depth of knowledge, or levels of proficiency, appropriate for each role and context” (DOL AI Literacy Framework, via GovCIO).

The National Science Foundation goes further. Its TechAccess: AI-Ready America initiative defines “AI-Ready” as “the continuum of literacy, proficiency, and fluency, the ability to understand, apply, and create with AI,” and it deliberately reaches beyond K–16 education to businesses, public-serving organizations, and individuals, emphasizing hands-on implementation, experiential learning, and coordination across local and state ecosystems (NSF TechAccess: AI-Ready America).

Together, these two anchors create the bridge for this work: the field already agrees readiness is a continuum that ends in creation. What it has not yet named is the human capability that makes creation durable. My argument is that readiness must include agency.

2. The Capacity Gap: AI Is Being Adopted Faster Than Human Capacity Is Being Built

AI is increasingly framed as a capacity tool or productivity tool, a way to do more with less. Microsoft’s 2025 Work Trend Index describes the emergence of the “Frontier Firm,” where work becomes AI-operated but human-led, organized around hybrid teams of humans and agents (Microsoft, 2025 Work Trend Index). The same report names a significant capacity gap: 53% of leaders say productivity must increase, while 80% of the global workforce reports lacking the time or energy to do their work, with employees interrupted every two minutes by a meeting, email, or ping (Microsoft, 2025 Work Trend Index).

Here is the overlooked point: AI is being adopted to close a capacity gap, but no one is yet measuring whether it actually builds human capacity or merely shifts the bottleneck.

Anthropic’s Economic Index offers an early signal that capacity is, in fact, learned. Its “Learning Curves” report finds that high-tenure users — those who signed up at least six months earlier — “have developed habits and strategies that allow them to better harness Claude’s capabilities,” attempting higher-value tasks and eliciting more successful responses, with roughly a 3–4 percentage-point higher success rate even after full controls (Anthropic Economic Index, March 2026). The report is consistent with learning-by-doing and even raises the possibility that the benefits of effective AI use are self-reinforcing (Anthropic Economic Index, March 2026).

Lasting AI fluency is not a result of first exposure. It develops through practice, patterns, and learning curves. A one-time workshop cannot manufacture what only repeated, self-directed use produces.

3. The Problem: Most AI Training Creates Exposure, Not Transfer

If capacity is learned over time, the dominant training model — the single workshop, the demo, the copied prompt — is structurally mismatched to the outcome it promises. This will not work for long-term AI fluency and enterprise AI adoption.

Here’s what we can gain from learning science. The National Academies define transfer as “the ability to extend what has been learned in one context to new contexts,” and they warn that some learning experiences produce memory but poor transfer, while others produce both (National Academies, How People Learn). Critically, they find that “knowledge that is overly contextualized can reduce transfer” and that “transfer across contexts is especially difficult when a subject is taught only in a single context rather than in multiple contexts” (National Academies, How People Learn).

This is exactly the failure mode of most AI training we see right now. A learner is shown one tool, in one use case, with one template. They remember the demo, but they cannot generalize.

If learners can only use AI inside the exact use case demonstrated in training, they do not yet have transferable fluency.

Attendance, satisfaction scores, and completed prompts measure exposure. They do not measure transfer. The gap between the two is where most AI investment quietly fails.

4. The Reframe: Literacy Is the Starting Point; Agency Is the Transformation

The progression is straightforward but rarely stated:

• AI literacy — understanding what AI is and how tools work.
• AI fluency — applying AI to real tasks.
• AI agency — the self-directed capacity to use AI with discernment, autonomy, judgment, creativity, and contextual intelligence across changing situations.

Literacy and fluency are necessary but not sufficient. The transformation organizations actually want — people who get more capable because of AI rather than more dependent on it — lives in agency.

I’ve been studying agency, and this reframe is grounded in self-determination theory, which holds that optimal human functioning depends on three psychological needs: autonomy, competence, and relatedness (Deci & Ryan, via TheoryHub). Translated into AI:

• Autonomy — Can I decide when and how to use AI, rather than being compelled by it?
• Competence — Can I use AI effectively and evaluate the result?
• Relatedness / context — Can I apply AI within my community, discipline, role, and lived reality?

Notably, self-determination theory also frames the autonomous choice to delegate — to “defer, delegate, or accept outcomes without coercion” — as an expression of agency, not a surrender of it (Self-Determination Theory review). That distinction is the heart of healthy human–AI collaboration.

5. Definition

AI agency is the self-directed capacity to use AI with autonomy, competence, discernment, and contextual judgment — initiating and sustaining AI use independently, evaluating outputs critically, and deciding what to delegate, co-create, or keep human.

Agency is not a soft skill. As the next section argues, it is responsible-AI infrastructure.

6. The Framework: The Agency Intelligence Stack™

My proprietary contribution to this research is the bridge between this research and daily practice. The Agency Intelligence Stack™ is a human capability model for the age of AI, sequenced from perception to standing:

Discernment → Taste → Creative Confidence → Critical Decision-Making → Agency → Authority

Mapped into the AI context, the stack becomes a diagnostic for what fluency actually requires:

Each capability maps to a recognized requirement. Discernment and taste correspond to the DOL framework’s own content area, Evaluate AI Outputs (DOL, via GovCIO). Critical decision-making and authority correspond to the human judgment that responsible-AI standards demand, as explored next.

7. Agency as Responsible-AI Infrastructure

The case for agency is not only developmental; it is a governance requirement.

NIST’s AI Risk Management Framework states that trustworthy AI is “valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed,” and that these characteristics must be balanced according to context of use (NIST AI RMF). Critically, the framework places human judgment at the center of deciding the metrics and thresholds for trustworthiness — these determinations cannot be fully automated in a user’s workflow.

Here’s the quotable for the industry:

Human agency is not a soft skill. It is responsible-AI infrastructure.

A workforce that cannot evaluate outputs, recognize quality, or decide what to keep human cannot operationalize trustworthy AI, no matter how well-governed the underlying systems are. Agency is the human layer that responsible AI assumes but rarely builds.

8. Measurement: Beyond Attendance and Satisfaction

If transfer is the missing outcome, then measurement must change. This research program proposes two instruments to move beyond attendance, satisfaction scores, and prompt completion:

• The AI Agency Outcomes Dashboard — tracking what learners do after training ends: whether AI use is initiated independently, transferred to new workflows, and sustained over time without a trainer or template.
• The AI Agency Capacity Score — a composite, mapped to the Agency Intelligence Stack™, assessing discernment, taste, creative confidence, critical decision-making, self-directed initiation, and trusted authority.

Both instruments are designed around a single principle drawn directly from learning science: measure transfer, not exposure (National Academies, How People Learn).

9. Research Agenda: Everyday AI Agency

The intention is to open a multi-study practitioner research program. The animating questions:

1. What happens after AI training ends?
2. Can participants transfer AI skills into new workflows without a trainer or template?
3. What distinguishes temporary AI exposure from lasting AI fluency?
4. How do users decide what to delegate, co-create, or keep human?
5. What conditions help AI increase agency rather than dependence?
6. How should institutions measure AI fluency beyond attendance, satisfaction, and prompt completion?

10. Implications

• Universities & HBCUs — Move from one-off AI workshops to multi-context, transfer-oriented curricula that build durable agency, not demo-day confidence.
• Workforce boards — Adopt outcome measures that track post-training transfer, aligning local programs with the NSF readiness continuum and DOL role-specific proficiency.
• Employers — Define role- and context-specific AI proficiency, and invest in the learning-by-doing conditions that the Anthropic Economic Index shows actually compound capability over time.
• Creators & solo founders — Treat agency as a competitive moat: discernment, taste, and creative confidence are what keep a human voice valuable in an AI-saturated market.
• Everyday knowledge workers — Reclaim autonomy by deciding when AI serves you and when it does not, turning a capacity tool into genuine capacity.

11. Call to Partner

The shift from AI literacy to AI agency cannot be studied from the sidelines. Jeneba Wint, Founding Principal and CEO of AI Atelier, invites academic partners, innovation labs, HBCUs, workforce programs, and funders to co-design the studies, instruments, and field pilots that will define Everyday AI Agency.

The national conversation has already moved from AI access to AI readiness. The next move — the one that determines whether AI builds human capacity or merely consumes it — is from readiness to agency.

Let’s build the evidence base together.

Sources

1. U.S. Department of Labor — AI Literacy Framework release: dol.gov
2. DOL AI Literacy Framework, coverage and quotations (GovCIO): govciomedia.com
3. NSF TechAccess: AI-Ready America (solicitation NSF 26-508): nsf.gov
4. Microsoft — 2025 Work Trend Index, “The Frontier Firm is born”: blogs.microsoft.com
5. Anthropic Economic Index — “Learning curves” (March 2026): anthropic.com
6. National Academies — How People Learn, Ch. 3 “Learning and Transfer”: nationalacademies.org
7. Self-Determination Theory (Deci & Ryan) — review: open.ncl.ac.uk
8. NIST AI Risk Management Framework — Characteristics of Trustworthy AI: airc.nist.gov