Requirements radar¶
English · Русский · AI Interview Atlas › Reference
- Period: 2026-09-15 — 2026-09-19
- Sample: 79 postings: engineering 17, leadership 62; markets: international 79, Russia 0, unknown 0
| Theme | AI Engineering | AI Leadership |
|---|---|---|
| LLM fundamentals | — | 15% (9) |
| Inference, serving and cost | — | 15% (9) |
| RAG and retrieval | — | 8% (5) |
| Agents, tools and protocols | — | 27% (17) |
| Evaluation and observability | — | 10% (6) |
| Safety, security and governance | — | 16% (10) |
| AI system design | — | 18% (11) |
| Practical coding | — | 29% (18) |
| AI product strategy and metrics | — | 21% (13) |
| AI platform and operating model | — | 13% (8) |
| Leading engineering teams | — | 48% (30) |
| Programmes and delivery | — | 82% (51) |
| Applied and customer scenarios | — | 13% (8) |
| Payments and regulated domains | — | 40% (25) |
Leadership postings in the sample: the role itself involves AI or machine learning in 42% (26 of 62).
A dash means fewer than 5 postings mention the theme, or the track has fewer than 20 postings; such cells are not published.
Method¶
79 public full-text job postings for senior product, engineering and programme roles, first seen 2026-09-15 – 2026-09-19 on employer career boards and job boards during one job search, deduplicated by employer and title. The track comes from the job title, with content checks to exclude functions outside scope; design, marketing, sales and similar functions are excluded. Claude Sonnet 5 classified the first 60 postings and GPT Astra completed the remaining batch against the criteria below, retaining evidence for every marked theme. GPT Astra checked a deterministic random sample of eight retained postings and the flagged ambiguous cases against the full source text. Author and review records are retained privately. A posting counts once per theme. “The role itself involves AI” means the role owns AI or machine-learning products, features, models, teams or infrastructure; using AI tools does not count. The sample leans towards international fintech and business software and is not a market survey. This edition has too few engineering postings to publish that track.
What counts for each theme¶
- LLM fundamentals: experience with or knowledge of LLMs, generative AI or foundation models
- Inference, serving and cost: AI/ML serving, inference or GPU infrastructure, or the latency and cost of AI workloads
- RAG and retrieval: retrieval-augmented generation, semantic or vector search, embeddings
- Agents, tools and protocols: AI agents, agentic workflows, AI tool use or agent protocols such as MCP
- Fine-tuning and post-training: training or fine-tuning machine-learning models
- Evaluation and observability: evaluating, testing or monitoring the quality of AI/ML systems
- Safety, security and governance: responsible AI, AI risk, AI regulation or the security of AI systems; general security or financial compliance without AI does not count
- Multimodal and voice: speech, vision or multimodal AI
- AI system design: designing the architecture of AI/ML systems or platforms; general software architecture does not count
- Practical coding: hands-on coding in the role or proficiency in a named programming language
- AI product strategy and metrics: product strategy, roadmap, metrics or the business case for AI/ML products or features; general product strategy does not count
- AI platform and operating model: internal AI platforms, AI adoption or transformation across an organisation, AI enablement
- Leading engineering teams: managing engineers or engineering managers: hiring, coaching, performance, team or organisation design
- Programmes and delivery: leading cross-functional delivery: roadmaps, planning, dependencies, launches and execution across teams
- Applied and customer scenarios: customer-facing technical deployment: forward-deployed, solutions or pilot work with customers
- Payments and regulated domains: payments, banking, lending, insurance, fintech or financial-regulation domain experience