The AI Competence Framework Landscape: A Comparative Analysis of Frameworks for Education and Workforce Development

Synthesis: This paper maps and compares institutionally endorsed AI competence frameworks for education, professional development, and workforce training, so that curriculum and training designers can choose among them on evidence rather than reputation. Fitsilis screened 76 candidates, shortlisted 24 frameworks and reference models, and applied eight inclusion criteria to bring 16 into a comparative content analysis. Each framework was coded in three blocks — profile, architecture, and competence coverage — and scored on a three-point scale across ten competence areas, from AI Foundations to Domain or Workflow Application. The comparison shows a shared core: AI Use and Interaction, Output Evaluation and Verification, and Domain or Workflow Application were explicit and substantive in all 16 frameworks, and AI Foundations and Ethics and Responsible AI in 15 each. Variation concentrates elsewhere. Problem Formulation, Data Competence, and AI Creation and Development were substantive in only seven, nine, and eight frameworks. The frameworks also differ sharply in whether they supply proficiency levels, learning outcomes, and Assessment guidance, or leave that translation to the adopting institution.

Key Findings

1. From 76 candidates, 24 frameworks were shortlisted and 16 met eight inclusion criteria for comparison; all 16 were published between 2024 and 2026 by international bodies, national authorities, universities, standards bodies, and consortia.
2. AI Use and Interaction, Output Evaluation and Verification, and Domain or Workflow Application were explicit and substantive in all 16 frameworks, forming the shared foundation of institutional AI competence.
3. Ethics and Responsible AI and AI Foundations were substantive in 15 frameworks each; Safety, Security and Risk in 14, and Human Agency and Oversight in 12, tying responsible practice to risk awareness and retained judgment.
4. Coverage was least uniform for Problem Formulation (7 of 16), AI Creation and Development (8 of 16), and Data Competence (9 of 16), the areas where general literacy frameworks and specialist professional frameworks diverge most.
5. Structure varies independently of content: some frameworks supply progression levels, learning outcomes, and pedagogical and assessment guidance; others leave teaching and assessment decisions to the institution.
6. Proficiency levels are not comparable across frameworks: some describe an individual learner's development, others the responsibilities of an occupational role, and curriculum frameworks grade-specific expectations.
7. The paper proposes AI bilingualism — turning an ambiguous human objective and domain knowledge into instructions and constraints an AI system can process, then interpreting its probabilistic output for human decision — as a competence future frameworks should make visible.

A shared core, a differentiated technical edge

The 16 frameworks agree most where practice is involved. Every one expects people to understand enough about AI to use it for a purpose, examine what it returns, and take responsibility for applying it in their own context. Safety, Security and Risk was substantive in 14 frameworks and Human Agency and Oversight in 12, so responsible competence is described through risk recognition and continuing human responsibility rather than ethical awareness alone. The divergence tracks audience. General literacy frameworks concentrate on interacting well with systems that already exist, while technical and professional frameworks for vocational education and workforce development also expect development, adaptation, and implementation of solutions. Data Competence, AI Creation and Development, and Problem Formulation were substantive in nine, eight, and seven frameworks respectively. Fitsilis reads the lower counts as role-specificity or as capabilities assumed from wider digital and lifelong learning, not as evidence that any area is becoming obsolete.

Structure decides how usable a framework is

A framework can specify the same broad competence as another and still give a teacher or training designer very different help. The architecture ranged from a few broad domains to hierarchical structures with subdomains, competence statements, proficiency descriptors, and learning outcomes. The educator frameworks specify knowledge, skills, and attitudes at different levels of development. UNESCO's frameworks set out competencies in four and five dimensions respectively and differentiate them over three developmental levels, while the UAE national curriculum framework organizes domains, strands, standards, and grade-specific outcomes for K–12 students. Progression models carry a trap: levels that look alike in a comparison table may describe individual learner development, occupational role responsibilities, or educational cycles, so a role-based ladder from a workforce framework does not show how one person progresses. Implementation support differs too — some frameworks provide formative and summative guidance and outcome statements aligned to Assessment, others leave that translation to the institution.

Five themes and the emergence of AI bilingualism

Read together, the ten areas resolve into five themes: foundational and systems literacy, applied interaction and human–AI co-creation, critical evaluation, responsible AI and AI Governance, and human-centered dispositions. The responsible-use theme covers Privacy and personally identifiable information, the boundaries of consent, academic integrity, and disclosure of AI assistance, with environmental cost noted by UNESCO and the AILit framework. Human-centered readiness keeps Self-Efficacy, reflection, and moral autonomy in view. Beyond the comparison, the screened corpus surfaced competencies concentrated in specialist documents: AI procurement and contracting, MLOps, agentic workflow orchestration, and AI-supported pedagogical design. Fitsilis argues the field is moving from literacy toward AI bilingualism — translating an ambiguous human objective into instructions and constraints a system can process, then interpreting its probabilistic output for human decision — a two-way competence spanning problem formulation, contextual judgment, and oversight.

What this means for practice

• Instructors. Check what a framework's proficiency levels describe before adopting them for a course: a role-based ladder does not map onto classroom progression, and a grade-specific curriculum ladder does not transfer to adult learners.
• Curriculum designers. Select on population, required competence level, and the task the learner will actually perform; choosing a framework because it contains a desired AI topic is not sufficient.
• Administrators and workforce developers. Treat problem formulation, data competence, and AI creation as deliberate additions for technical or specialist programs; they were substantive in only seven, nine, and eight of the 16 frameworks.
• Researchers. The ten-area scale is a comparison instrument, not another competence framework; the paper calls for work on adoption, validated assessment instruments, and how AI bilingualism can be taught and assessed. See interpreting and applying AIED research.

Limitations

• The analysis covers only institutionally endorsed frameworks with sufficient public documentation; independent academic proposals that did not meet the eight criteria were excluded.
• The search ended in September 2026, so later publications or revisions may alter the landscape described.
• The three-point coverage scale required interpretive decisions, even though the analytical areas were defined in advance.
• The study examined the content of framework documents, not their application in education or organizations; no adoption or outcome data were collected.

Citation

Fitsilis, P. (2026). The AI Competence Framework Landscape: A Comparative Analysis of Frameworks for Education and Workforce Development. EdArXiv preprint.