AI Engineer (Role)
AI Engineer (Role)
The job of building systems around existing models — wiring frontier or open models to data, tools, memory, loops, and guardrails so they do useful work. Not to be confused with AI Engineer, the conference series/channel.
AI engineer vs ML researcher
Per Cedric Clyburn in Essential Skills for Becoming an AI Engineer (IBM Technology):
| ML researcher | AI engineer | |
|---|---|---|
| Builds | Foundation models from scratch; new architectures | Applications/systems on top of existing models |
| Output | Papers, models | Working solutions users can actually use |
| Typical bar | Deep math, usually an advanced degree | Fundamentals + judgment; no CS degree required |
| Metaphor | Builds the engine | Builds the car |
Demand sits overwhelmingly on the car-builder side: "we desperately need more folks who can build these cars and AI solutions for organizations."
Skill stack — entry view (Clyburn, three tiers, in order)
- Foundations — Python (to read what agents write), Git / CLIs / Linux, APIs (rate limits, response handling).
- AI engineering — embeddings & vector search, RAG, agents & tool use (Agentic Loop).
- Ship & deploy — containers & Kubernetes, observability, monitoring (token spend, security).
Skipping tier 1 to jump to agents is the most common failure mode — learners end up relearning basics.
Skill stack — production view (Kopecki, seven disciplines)
From The 7 Skills You Need to Build AI Agents (IBM Technology): system design · tool & contract design · retrieval engineering · reliability engineering · security & safety · evaluation & observability · product thinking. Kopecki calls the resulting role the "agent engineer" — see Agentic Engineering.
Together these read as an entry → senior progression: Clyburn's tiers get you in; Kopecki's disciplines are what make agents survive production.
Why the role exists now
- Code generation is cheap; judgment (what to build, how to structure it, why one approach over another) is the scarce input — see Code Is Free.
- Most of the senior disciplines are re-pointed backend/distributed-systems/security skills, which is why redeploying platform/SRE/security engineers is often faster than hiring — see Designing IT Roles for an AI Era (Talent Strategy POV) and Skill Change Index (SCI).
- Fundamentals matter more, not less, when agents write the code — see Software Fundamentals Matter More Than Ever (Matt Pocock, AI Engineer).
Open tension
"Learn by building" with AI tools is the recommended on-ramp, but if the agent does the building the learner may never acquire the judgment the role is defined by — the Deskilling Trap (Juniors). Clyburn's "Python fluent enough to read what your agent writes" is a partial answer: comprehension as the minimum bar.
Sources
- Essential Skills for Becoming an AI Engineer (IBM Technology) — role definition, engine/car split, three-tier stack, top production use cases
- The 7 Skills You Need to Build AI Agents (IBM Technology) — seven production disciplines