What this model is
The EMA AI Competency Model names the six competencies the EMA AI Extension examination assesses. It is published in full so that a candidate knows what is examined and an employer knows what the extension certifies.
It is a standard, not a syllabus, on the same terms as the EMA Competency Framework: it says what capability looks like, not how to acquire it. The Academy sells no training and endorses none.
Three documents make up EMA AI Standard v1.0 and are released together:
- This model: the six competencies and their permanent IDs.
- The EMA AI Body of Knowledge: one unit per competency, written for judgment rather than recall.
- The EMX-AI Exam Content Outline: for every competency, how many questions it carries, what a certified holder can do, and what kinds of situations the questions present.
Why it is a separate standard
AI practice changes faster than the rest of engineering management. The evidence on productivity is contested and moving, the regulation is new, and the tools a team uses this year may not exist in two. The core standard promises that its competencies change at most once a year, and that promise is worth keeping. So the AI Standard is versioned on its own, with its own release cadence, and the extension it examines is valid for two years rather than three. The governance chapter states the terms.
How it relates to the core
The extension is added to a ladder credential, not placed on the ladder. A candidate must hold a valid EMA-I, EMP-II or EME-III when the sitting starts.
That prerequisite does real work. Competency TJ-5 of the core framework, AI-assisted engineering, is the baseline AI judgment every engineering manager needs: introducing assistance with clear expectations, recognising its risks, telling real help from looking fast. Every ladder credential already certifies it. The AI Standard assumes that baseline and goes deeper, into the decisions of a manager who leads an organisation's adoption of AI, or a team that builds and runs AI-powered products.
The boundary rule. TJ-5 is the baseline. The AI Standard goes deeper and never contradicts it. Where a release of the AI Standard would contradict TJ-5, TJ-5 is corrected in a minor version of the core standard in the same quarter, and both changelogs record it.
The domain: AI in engineering leadership
Leading the adoption, delivery and operation of AI in engineering, so that it earns its place, stays accountable, and keeps the team able to understand what it ships.
| ID | Competency | One line |
|---|---|---|
| AI-1 | Adoption strategy and staged bets | where AI earns its place, sizing a bet to teach, and when to stop |
| AI-2 | Governance, security and compliance | data flows, injection and agency risks, proportionate policy, regulated uses |
| AI-3 | AI-assisted software delivery | accepting generated code, agent permissions, review and testing when code is cheap |
| AI-4 | Building AI-powered products | choosing the approach, evaluations, non-determinism, cost and model dependencies |
| AI-5 | Operating AI systems in production | quality and cost signals, model changes as production changes, AI incidents |
| AI-6 | Capability, roles and measuring impact | building skill, protecting early careers, assessing people, measuring honestly |
AI-3, AI-4 and AI-5 are technical. They carry 34 of the examination's 60 questions, and they still test a manager's judgment about a technical decision, never engineering recall. A candidate is expected to read an architecture sketch, question an evaluation result, or decide whether an agent may merge its own changes. No question asks them to write code, a prompt or a configuration, or to name a product.
What the model does not cover
- The baseline in TJ-5. The prerequisite certifies it.
- Machine-learning theory. Training methods and model internals appear only where a manager needs the mechanism to judge a decision.
- Legal interpretation. The model covers when to involve legal or compliance, not what an article of a regulation requires.
- Products, vendors and their current figures. These change faster than any standard can.