On 18 June, in Brussels and online, the European Commission and the OECD launch the final version of their AI Literacy Framework. It's a common, global account of what young people need in order to understand, create with, manage, and question artificial intelligence. Supported by Code.org and feeding into the PISA 2029 assessment, it's serious work. And it arrives in a crowded field.
A single research tool at the Hong Kong University of Science and Technology already catalogues 20 AI literacy frameworks, spanning K–12 standards, higher-education competency models, workforce programs, and clinical training. The reason that tool had to be built is the tell: different frameworks name the same skills differently, so someone had to construct a translator before the documents could speak to each other. The content questions are largely settled. The open question is which of these frameworks actually gets used.
In a field this crowded, what separates a framework that travels from one that sits on a shelf is rarely its quality. It's how easily the systems where learning and hiring happen can pick it up and put it to work. Those systems (learning management systems, credentialing platforms, job boards, and hiring tools) run on structured data: stable identifiers, importable taxonomies, and machine-readable formats. A framework published only as a document asks every institution and vendor to do that integration itself, one interpretation at a time. Publish the same framework as structured data, and the work is done once, for everyone.
The new framework targets schools. But the competencies a student builds there only matter if they stay legible as that learner moves into higher education and work, and that hand-off is exactly where mismatched frameworks break down. A PDF is a photograph of a framework. You can read it. You can't run it.
Where machine-readability already works.
Europe has the clearest example in its own infrastructure. ESCO, the EU's multilingual classification of skills and occupations, covers 27 member states and was published from the start as linked open data: stable identifiers for every element, downloadable in standard formats, free for any platform to integrate. Labor platforms, learning systems, and credential systems across Europe adopted it, not because anyone required them to, but because integration was easier than building from scratch. The content mattered. The format made the content usable.
The United States built the same thing for occupations. ONET, the federal Occupational Information Network, maps skills, tasks, and knowledge across hundreds of jobs, and it's published as linked open data, too. The result is that ONET is embedded across job boards, workforce tools, and labor-market analytics nationwide. No mandate. Just a format that made adoption the path of least resistance.
This is the fork every framework reaches. Without a machine-readable layer, a framework depends on manual interpretation across thousands of institutions and vendors, which produces fragmented alignment at best. At worst, it becomes a "standard" that each platform reads differently, until comparability, the entire point, is gone.
Where adoption actually happens.
Once a framework is structured, the systems that reach learners can do something with it.
In a learning management system, native alignment means an instructor doesn't re-interpret the framework for every course. Outcomes get tagged the same way across programs, analytics become comparable, and when the framework updates, connected courses update with it. The work of alignment happens once, in the platform, instead of thousands of times, by hand.
In a credentialing system, the identifier is what makes a credential portable. A badge that carries the framework's competency identifiers is legible to employers and recognizable to hiring systems anywhere. A badge that carries only a label, "AI Literacy Certificate," means something different at every issuer, and the shared understanding the framework promised dissolves into noise.
None of this requires a framework to be complex. It requires it to be structured: each competency assigned a stable identifier, the full taxonomy available in importable formats, and explicit crosswalks to the infrastructure that already exists, ESCO and O*NET, plus the other frameworks already in use.
The cost of skipping this is quiet fragmentation.
Without machine-readability, the pattern is predictable. Institutions implement by hand, or not at all. Vendors build their own interpretations. Credentials multiply without comparability. Employers stop trusting the signal. The framework becomes a reference the field cites but doesn't systematically use. With this many frameworks already in circulation, each new PDF doesn't reduce the confusion. It adds another dialect to it.
This isn't a failure of content or intent. It's a failure of format, and it lands hardest where capacity is thinnest. Well-resourced universities and large employers can interpret dense guidance by hand, and they will. Community colleges, workforce programs, and apprenticeships, which serve the learners who most need a portable AI literacy credential, are left to improvise. Machine-readability is what turns a document the strong can implement into infrastructure everyone can adopt.
What the next step looks like.
The ask is concrete and well within reach. Every framework author, the EU-OECD initiative first among them given its reach, should publish a structured implementation package alongside the PDF: a machine-readable version in which each competency carries a stable identifier, the full taxonomy is available in standard formats, and explicit crosswalks connect it to ESCO, O*NET, and the major frameworks already in use.
This needs no new program and no significant funding. It means treating a framework as public data infrastructure, which is exactly what ESCO and O*NET already are.
We don't need another framework. We need the ones we already have to be usable by the systems that decide whether a learner's AI literacy ever counts. The content is written. The infrastructure step is next.
