Essential Skills for Becoming an AI Engineer: RAG, AI Agents, & More

Summary

In this ~11-minute IBM Technology explainer, Cedric Clyburn defines the AI engineer role and lays out a three-tier skill stack for getting into it. His opening premise: AI coding tools have made generating code easy, so "the code itself isn't the hard part anymore, the hard part is judgment" — knowing how to structure applications, what to build, and why one approach beats another. That judgment, he argues, isn't taught in a class; it's learned by building.

He draws a sharp line between the ML researcher (trains foundation models from scratch, publishes papers, needs deep math and usually an advanced degree) and the AI engineer (builds with existing frontier or open models and wires them into systems that do useful work — data, tools, memory, loops, guardrails). "If ML researchers are the ones building the engine, then AI engineers are the ones building the car." And right now, with the influx of models, organisations "desperately need more folks who can build these cars." No CS degree required — but fundamentals are.

The three-tier skill stack (order matters)

Clyburn's key sequencing claim: people skip tier one and jump straight to building or deploying agents before they can handle data or understand infrastructure — "so they spend a lot of time relearning the basics."

Tier	Focus	Skills named
1. Foundations	Not AI-specific, but you can't build without it	Python (fluent enough to read and understand what your AI agent is writing, not a wizard); Git, CLIs, Linux (the OS your agents deploy onto); APIs (programmatic model calls, response handling, rate limits — "every AI product… is fundamentally well-structured API calls")
2. AI engineering	The AI-specific core	Embeddings + vector search (meaning over keyword match; e.g. "Kubernetes" near "containers", "orchestration"); RAG (ingest → fixed-size chunk → embed → store; at query time retrieve + stuff into the context window → grounded answer); Agents & tool use ("probably the most in-demand applied AI skill right now")
3. Ship & deploy	Get it off the laptop and into users' hands	Containerization + Kubernetes (package agent, sometimes the model, across hybrid cloud); Observability (why did the agent make that decision — "transparency and trust"); Monitoring (token-bill spend, security)

Workflow vs agent (his quick distinction, see Agentic Loop): a workflow follows a predefined path (A → B → C); an agent dynamically decides what to do next — calls a tool, observes the result, loops. "One who can do this reliably and at scale, well, that's a good AI engineer."

What's actually in production (his read)

The three most common AI-in-production use cases he sees:

RAG knowledge systems — HR services, hospitals, customer chatbots, internal Q&A over company data. "Almost every company experimenting with AI wants some version of RAG."
Agents with tools — querying databases, visualising data, doing things "that typically a subject matter expert only could."
AI-accelerated shipping — helping engineers get code shipped "in hours and not weeks."

His career advice: build portfolio projects in those three areas, aligned with your own interests.

note Gap in the source The intro promises "three projects that'll help demonstrate your skills to potential employers"; the video actually delivers three production use-case categories and a generic "build in these areas" nudge, not specified projects. Treat it as a skills-stack map, not a portfolio curriculum.

Key takeaways

Judgment, not code, is the scarce input — a third independent voice for the Code Is Free thesis, this time pitched at career-entrants rather than frontier teams.
Python literacy is now reading literacy. The bar is "understand what your agent is writing," not "write it from scratch" — a concrete reframing of what foundational skill means when agents author the code.
Sequence matters: foundations → AI core → ship. Skipping tier 1 is the most common failure mode he sees — corroborates Software Fundamentals Matter More Than Ever (Matt Pocock, AI Engineer) from the entry-level side.
RAG is still the enterprise default workload, agents are the hottest skill, and deployment/observability is where demos die.
AI engineer ≠ ML researcher. The hiring pool for the car-builders is far larger than for the engine-builders — and it doesn't require an advanced degree.

How it relates to Kopecki's "7 Skills"

The 7 Skills You Need to Build AI Agents (IBM Technology) (same channel, June 2026) is the depth view — seven production disciplines for agent engineers already on the job. Clyburn's is the on-ramp view — what order to learn things in to become one. They overlap on retrieval, agents/tools, and observability; Clyburn adds the tier-1 plumbing (Python/Git/Linux/APIs) and deployment (containers/Kubernetes) that Kopecki assumes; Kopecki adds reliability engineering, security/safety, tool-contract design, and product thinking that Clyburn barely mentions (security appears only as a monitoring aside). Read together, they form a reasonable entry-to-senior progression for the AI Engineer (Role).

Cross-vault relevance

tip A ready-made curriculum ladder for an IT upskilling program. The three tiers map directly onto a capability-building sequence: tier 1 is the gate (many enterprise IT staff — analysts, service managers, run-ops — lack Python/Git/API fluency and that's where upskilling should start, not with "build an agent"), tier 2 is the AI core, and tier 3 is where platform/SRE engineers already have an advantage. The "Python fluent enough to read what your agent writes" bar is a useful, lower and more achievable target for non-developer IT roles than traditional coding bootcamps. Combined with Kopecki, it supports the talent-strategy conclusion: redeploy and upskill existing engineers rather than hunting for scarce "AI" hires — and the ML-researcher/AI-engineer split says most enterprises need car-builders, not engine-builders. Watch the Deskilling Trap (Juniors) tension: "learn by building" with AI tools only builds judgment if the learner actually understands what the agent produced.

Wikilinks

IBM Technology · Cedric Clyburn · AI Engineer (Role) · RAG · Agentic Loop · Code Is Free · The 7 Skills You Need to Build AI Agents (IBM Technology) · Software Fundamentals Matter More Than Ever (Matt Pocock, AI Engineer) · Designing IT Roles for an AI Era (Talent Strategy POV) · Deskilling Trap (Juniors) · Agentic Engineering · Context Engineering