The Frameworks Are Falling Short
What the AI Fluency Conversation Is Missing
Organisations everywhere are declaring AI fluency a strategic priority. Billions are being spent on training, certification, and upskilling programmes. The urgency is real. What is less often questioned is whether the foundations those programmes are built on are fit for purpose. There is a phrase being repeated in boardrooms, HR decks, and conference keynotes right now: "We need to build AI fluency across our organisation." But here is the question almost no one is asking: Fluency for whom?
This post examines the major industry frameworks that currently define how AI fluency is understood, measured, and taught. Each one has genuine merit. Each one also carries blind spots that, at scale, will systematically exclude the people who need this most.
What We Mean by AI Fluency
There is a meaningful difference between AI literacy and AI fluency. Literacy means understanding what AI is. Fluency means knowing how to work with it effectively, critically, and responsibly in your specific context. The most widely adopted individual framework is the 4D model, developed by Professors Rick Dakan and Joseph Feller in collaboration with Anthropic. It defines fluency across four competencies: Delegate (knowing what to hand to AI), Describe (articulating what you need clearly), Discern (evaluating what comes back), and be Diligent (using AI ethically and safely).
It is a solid starting point. The problem is that virtually every major framework in this space was built with a particular kind of person in mind: English-speaking, neurotypical, desk-based, digitally comfortable, employed in a mid-to-large organisation in the Western world. Most people do not fit that description, and no framework has fully reckoned with that fact.
Some of the Major Frameworks and Where They Fall Short
Earlier, we briefly touched fluency framework built collaboratively with Anthropic. Some other known industry frameworks are:
1. Google AI Professional Certificate / Google AI Essentials
Google's programme is practical and accessible. It builds a collaborator mindset, teaching learners to give AI clear instructions so it functions as a professional partner rather than a simple task tool. For someone new to AI, it is a reasonable entry point.
The limitation is structural. It is, fundamentally, a product fluency programme. The collaborator mindset it teaches is defined in terms of Google's ecosystem. There is no assessment of where a learner is starting from, no pathway differentiation by role or industry, no language accommodation, and no consideration of different learning needs or accessibility requirements. Fluency, as Google defines it, means comfortable and productive use of Google AI products. That is not the same thing.
2. Microsoft AI Skills Initiative / Career Essentials in Generative AI
Microsoft has made a significant investment here. Its programme offers role-specific training paths across business users, technical implementers, and strategic leaders, with over 15,000 professionals completing certification in Q1 2026 alone. The LinkedIn Learning partnership has given it broad reach.
The role categories, however, are too coarsely defined to be practically useful. "Business user" covers everyone from a warehouse supervisor to a hedge fund analyst. The programme is also tightly integrated with Microsoft's product stack, which makes it largely inaccessible to organisations that do not run on Microsoft 365. Microsoft's own Global AI Adoption research acknowledges that AI adoption in the Global North grew almost twice as fast as in the Global South in 2025, a gap its training design does not meaningfully address. Older workers, non-English speakers, and differently-abled learners are not considered.
3. UNESCO AI Competency Frameworks
UNESCO's frameworks for students and teachers are among the most philosophically ambitious in this space. The Teacher framework outlines 15 competencies across five dimensions, including AI ethics, pedagogy, and professional learning. It positions learners as active co-creators of AI rather than passive users.
The challenge is implementation. Contextual adaptability is the framework's most significant weakness: it was developed by researchers and policymakers from well-resourced educational systems, and applying it in low-resource, multilingual, or under-digitised contexts requires substantial local adaptation that the framework itself does not provide. It is also education-focused, and the transition into workforce application is left unaddressed. While the ethics dimension is strong in principle, the framework does not operationalise inclusion for learners with disabilities, neurodiverse learners, or speakers of languages outside UNESCO's working languages.
4. EU DigComp 3.0
Released in November 2025, DigComp 3.0 is a substantial update to the European Commission's digital competence framework, now incorporating AI literacy alongside cybersecurity, wellbeing, and sustainability. The context is striking: 92% of EU workers now use digital technologies in their jobs, 30% use AI systems, and yet only 15% have received any AI training.
DigComp 3.0 is a reference framework, not a delivery mechanism. It defines what competencies exist rather than how to build them, at what pace, or for whom. It is built around an imagined average European citizen that systematically underrepresents Europe's 87 million people with disabilities, 67 million non-native speakers, and significant socioeconomic variation across member states. In 2023, only 56% of EU adults had basic digital skills, meaning the framework assumes a foundation that nearly half the population has not yet established.
5. OECD-EC AILit Framework
Released as a draft in May 2025, this joint OECD and European Commission framework is the most comprehensive attempt to define global AI literacy standards for school-age learners. The expert group that developed it included representatives from Canada, the USA, Uruguay, Germany, France, and Croatia.
The absence of Sub-Saharan Africa, South and Southeast Asia, and the Arabic-speaking world from that group is not a footnote. Those regions account for the majority of the world's school-age population. The framework is well-constructed for the contexts its developers know well, but its claim to global relevance is undermined by that representational gap. It is also explicitly scoped to children and educators, leaving the adult workforce undergoing AI transformation right now entirely outside its remit.
6. Alan Turing Institute AI Skills for Business Competency Framework
The Turing framework is the most workforce-focused of the six. It defines competencies across three groups, General Employees, AI Professionals, and AI Leaders, developed through consultation with UK employers across multiple sectors. Version 3 was released for national consultation in December 2025.
It is a UK framework, shaped by UK conditions, and that scope limits its wider applicability. Academic critique has noted a bias toward AI algorithm literature and a reliance on specialised case studies without broader systematic research across diverse learner populations. The three-tier role structure is too broad to guide practical upskilling: "General Employee" encompasses a hospital porter and a marketing director with equal imprecision. Neurodivergent workers, differently abled learners, and the significant linguistic diversity of the UK's own workforce are not addressed.
Five Structural Limitations Across All These Frameworks
They define the problem, not the person. Every framework defines what AI fluency involves. None adequately accounts for the diversity of people who need to develop it, their different starting points, psychological states, cognitive profiles, linguistic contexts, and rates of learning.
They are designed for the already-connected. Docebo's 2026 AI Readiness Gap report found that 85% of employees say the training they receive does not help them use AI in their role, and nearly 60% feel their organisation's learning programmes are not designed with people like them in mind. Frameworks built for organisations with established L&D infrastructure will not reach the people furthest from fluency.
They treat ethics as a module rather than a mindset. Every framework includes an ethics dimension. In every case, it is framed as knowledge to acquire. None of them address the human cognitive biases that continue operating after the ethics module is complete: sycophancy acceptance, automation trust, and confirmation loops are documented and consequential, and they are absent from the curriculum.
They are geographically and linguistically narrow. Even the frameworks with global mandates were built by teams drawn predominantly from high-income, English-literate, Western systems. The Global South, non-English-speaking workforces, and low-resource contexts are acknowledged as concerns but are not reflected in the actual design choices.
They were not built with differently-abled people. The American Foundation for the Blind's 2026 research states it plainly: students and workers with varying-needs, who stand to gain the most from AI tools, are often the most disadvantaged when accessing them, and few differently-abled people have been asked to advise on the development of these products.
The Gaps the Frameworks Do Not Name
The frontline worker is invisible. Approximately 80% of the global workforce, around 2.7 billion people, are frontline workers in manufacturing, healthcare, retail, logistics, and agriculture. Building AI fluency for the deskless, the shift-based, and the care-sector workforce requires voice-first interfaces, context-specific training, supervisor capability building, and cultural change. None of these appear in any standard framework.
The language gap is structural, not cosmetic. Every major framework was developed in English. Most major training resources are delivered in English. The AI tools those resources teach people to use were predominantly trained on English-language data and perform better, more accurately, and more safely in English than in most other languages. For a worker who is a domain expert but English-limited, the gap between what they know and what they can express in a prompt gets mislabelled as a fluency gap. It is a language gap. Treating them as the same produces training that reliably fails the people who need it most.
The cognitive bias problem is being ignored. In controlled studies, a 2026 peer-reviewed study in Science found that AI affirmed users' actions 49% more often than human advisors did, even when those actions involved harm or deception. The International AI Safety Report 2026 documents that clinicians' diagnostic accuracy declined after extended AI-assisted practice. Research also shows that higher AI literacy can intensify biased processing rather than reduce it.
Differently abled workers are an afterthought. People with disabilities represent 16% of the global population and are systematically underrepresented in employment. AI has genuine transformative potential here, yet none of the frameworks have designed around these realities. The training delivery itself, typically video-heavy courses, timed assessments, and live webinars, frequently excludes the very people who would benefit most from the content. Neurodivergent workers may be the most overlooked group in this conversation, despite being among those with the most to gain. Research shows neurodivergent professionals in inclusive environments achieve a 20% increase in proficiency in AI and data skills. People with ADHD, autism, and dyslexia are already self-discovering AI as a profound cognitive equaliser. Fluency frameworks, however, are not designed for how they learn, process, or interact, so in the absence of intentional design they are left to figure it out alone.
The manager layer is broken. The 2026 State of Learning for AI Fluency Report found that 77% of executives believe their managers are prepared to guide employees through AI capability development. Among individual contributors, 91% say their managers are not fully prepared to do this.
The role and level disconnect is unresolved. Executives need the judgment to evaluate a multi-million-pound AI implementation. Individual contributors need practical workflow integration. Middle managers need to translate strategy into team practice. The vast majority of AI fluency training treats all three as if they need the same content at the same pace.
So What?
The frameworks are not wrong. They are incomplete. They were built in good faith by people working within the contexts they knew, and they have moved the conversation forward in important ways. The problem is that at scale, incomplete frameworks produce incomplete outcomes. Organisations following the mainstream playbook will build AI fluency for the people already closest to fluency. Everyone else will be left further behind, in programmes that were never really designed for them. The question is whether we are willing to redesign.
In the next post, we set out a different approach: how AI fluency development should be built, what a human-centred maturity model looks like in practice, and where the organisations getting this right are already doing differently.
Sudarshan's spent two decades in enterprise transformation, strategy and governance, working across platforms and industries to accelerate tech modernisation for global clients.
Born in India and formerly Global CTO and Cloud Lead at a major consulting firm, he specialises in data and AI — and in having the difficult conversations clients usually avoid, especially the ones about what not to do.
References
1. Dakan, R. & Feller, J. (2025). Framework for AI Fluency, Version 1.1. Ringling College / University College Cork. aifluencyframework.org
2. Google (2025). Google AI Professional Certificate. coursera.org/professional-certificates/google-ai
3. The Interview Guys (2026). Essential AI Skills for the Modern Workplace. blog.theinterviewguys.com/essential-ai-skills
4. Microsoft (2025). New Future of Work Report 2025. microsoft.com/en-us/research
5. UNESCO (2024). AI Competency Framework for Teachers. unesco.org/en/articles/ai-competency-framework-teachers
6. Mutawa, A. M. & Sruthi, S. (2025). UNESCO's AI Competency Framework: Challenges and Opportunities. IGI Global. doi.org/10.4018/979-8-3693-0884-4.ch004
7. European Commission / JRC (2025). DigComp 3.0. joint-research-centre.ec.europa.eu
8. OECD & European Commission (2025). Empowering Learners for the Age of AI. ailiteracyframework.org
9. Alan Turing Institute et al. (2025). AI Skills for Business Competency Framework, Version 3. turing.ac.uk
10. CIES (2025). Designing a Conceptual Framework of AI Capital. convention2.allacademic.com/one/cies/cies25
11. Docebo (2026). The AI Readiness Gap. businesswire.com/news/home/20260407928752/en
12. American Foundation for the Blind (2026). The AI Quagmire. afb.org
13. Acorn (2026). 2026 State of Learning for AI Fluency Report. businesswire.com/news/home/20260518188991
14. Sharma, M. et al. (2026). Sycophantic AI decreases prosocial intentions. Science. science.org/doi/10.1126/science.aec8352
15. International AI Safety Report (2026). gov.uk/government/publications/international-ai-safety-report-2025
16. The Conversation (2025). AI's Fluency in Other Languages Hides a Western Worldview. theconversation.com
17. Anthropic (2026). The AI Fluency Index. anthropic.com/research/AI-fluency-index
18. McKinsey & Company (2025). The State of AI 2025. mckinsey.com
19. World Economic Forum (2025). Future of Jobs Report 2025. weforum.org
