AI roles: responsibilities, interviews and preparation¶
English · Русский · AI Interview Atlas › Tracks and roles
Choose a role by the work you want to own. AI hiring combines established professions with changing specialisms; these six families are not six newly invented occupations. Titles overlap, and an “AI engineer” may build customer applications, operate infrastructure or conduct experiments. The useful distinction is the deliverable, the decision authority and the failure the person must resolve.
Each role page combines three kinds of material that carry different weight. What the role involves summarises public job postings: they describe responsibilities at a few employers, mostly frontier-model developers, so they illustrate a role rather than survey a market, and a posting never counts as evidence that a question was asked. How it is interviewed and the reported questions come from company pages and question evidence, with the usual markers. Practice questions and the preparation route are editorial recommendations.
Job postings checked on 2026-09-29. They describe responsibilities, not how candidates are interviewed.
Markers: ✅ confirmed by the company · 🗣 candidate report · † prep guide or compilation without a first-hand source · 🧪 generated from job-posting themes. Interview loops change often, so check the retrieval date.
| Role family | What it owns | Reported stages | Companies | Reported questions | Practice questions |
|---|---|---|---|---|---|
| Forward Deployed Engineer (FDE) | Gets a useful AI system working in a customer's environment, from discovery to handoff. | 17 | 5 | 14 | 14 |
| Applied AI / agent product engineer | Turns model capabilities into working product behaviour: interfaces, agent runtimes, tools and recoverable failures. | 12 | 5 | 9 | 18 |
| AI evaluation and reliability engineer | Produces trustworthy evidence about system behaviour: measurement strategy, regressions and release decisions. | 0 | 0 | 0 | 14 |
| Inference / AI platform engineer | Makes AI computation and shared services perform: bottlenecks, validated optimisations and telemetry. | 4 | 1 | 1 | 18 |
| AI product leadership | Owns defensible product decisions and measurable customer value for AI products. | 0 | 0 | 3 | 11 |
| Research engineer | Combines experiments with the software that runs them: architectures, data, training and analysis. | 0 | 0 | 0 | 12 |