Path 1 of 10, by 2026 demand
AI Engineer
Build LLM-powered products in industry: retrieval, tools, agents, evals and the production engineering around them.
Demand #1 fastest-growing US role on LinkedIn for the second year running (third-party, Sources M1). Evals and agents appear in roughly 40 to 50% of lab engineering posts (our count, §2.3).
What the role is
An AI Engineer builds product features on top of foundation models: assistants, copilots, document processing, agents. The work is mostly software engineering: APIs, data pipelines, evaluation harnesses, observability and cost control, with the model as an unreliable but powerful component.
It is not ML research. Training models from scratch is rare; prompting, retrieval, tool design, fine-tuning occasionally, and above all measurement are the core skills. LinkedIn's most-cited AI Engineer skills are LangChain, RAG and PyTorch (third-party, Sources M2).
Level signals
- Senior: ships an LLM feature end to end with an eval set and monitoring.
- Staff: sets the evaluation and release process for several teams; chooses between RAG, tools, fine-tuning and workflow designs with stated trade-offs; owns cost and latency budgets.
- Principal: defines the company's AI platform and governance (gateway, model policy, data handling) and the operating model for AI features across the org.
The typical loop
Loops vary by company. This is the common shape in third-party reports and job posts; confirm with the recruiter.
| Stage | What it tests | Basis |
|---|---|---|
| Recruiter and hiring-manager screens | Motivation, recent hands-on LLM work | Common industry pattern; no single source |
| Coding screen | Data structures in Python, sometimes an LLM API task | Python default reported for lab loops (third-party, A9) |
| LLM system design | RAG, agents, evals, cost and failure handling | Common industry pattern; no single source |
| Practical / take-home | Build a small LLM feature with tests and an eval | Common industry pattern; no single source |
| Project deep dive | A real LLM project: decisions, numbers, what broke | Common industry pattern; no single source |
| Behavioural | Ownership, collaboration, judgement | Common industry pattern; no single source |
Track order
Work top to bottom. Ticks come from the lessons you mark complete.
- 01Building blocksLLM and agent engineering track. The core of the role; start here.
- 02Production AI systemsLLM and agent engineering track. Agents, evals, guardrails, cost: where interviews go deep.
- 03Arrays and stringsCoding patterns track. Python fluency for the coding screen.
- 04Searching and orderingCoding patterns track. The rest of the most-asked patterns.
- 05The design methodSystem design track. The design method you will apply to LLM systems.
- 06Worked promptsSystem design track. Two of the four worked prompts are LLM systems.
- 07Securing AI systemsSecurity engineering for AI systems track. Injection and data handling come up in every LLM design round.
- 08Behavioural rounds at Staff levelLeadership and behavioral track. Story bank with at least two LLM delivery stories.
- 09Mock roundsMock rounds track. Coding, design and project deep-dive mocks.
Role practice set
Prompts written for this site. They are not reported interview questions.
- Build a RAG assistant over a public document set (for example your cloud provider's documentation) with a 50-question golden set, and report recall@5 and faithfulness.
- Design an LLM feature that classifies and routes support tickets; include the eval, the fallback when the model is down, and the monthly cost at 100k tickets.
- Take an agent that loops; add budgets, tracing and a human approval step, and show the before-and-after on cost and success rate.
- In 45 minutes, design the multi-tenant LLM gateway from the design track and defend the fallback policy.
- Explain to a skeptical architect when fine-tuning beats retrieval, and when it does not.
Your résumé and story angle
- Lead with "AI integrations" as production engineering. [Project]: what you integrated, for how many users, how you measured quality, what it cost. If you have no eval story yet, build one this month (practice item 1).
- Distributed systems are your differentiator. Agents are long-running, failure-prone workflows; say that you have built durable, observable systems at scale on Azure and Kubernetes. [Numbers.]
- Show Python. The labs post Python in about half their engineering roles and C# in none (our count, §2.3). Put a recent Python project on GitHub and on the résumé.
- Architect credibility. You can explain security, identity and data residency to enterprise stakeholders; that is rarer than prompt skill.
Six-week plan
Tick weeks off as you go
Sources
- Third-party LinkedIn: Jobs on the Rise 2026 press release (2026-01-14)
- Third-party Dice: summary of LinkedIn's 2026 list (2026-01-14)
- Third-party Indeed Hiring Lab: AI postings vs. tech postings (2026-01-22)
- Third-party Aced (Exponent): Anthropic interview process
- Reference Full research notes: docs/RESEARCH.md