AI Engineer Roles
“AI Engineer” covers a wide range. Different roles emphasize different parts of this learning path. Knowing which role fits you focuses your effort.
The role spectrum
1. AI Application Engineer / AI Product Engineer
What they do: ship AI features in user-facing products. Mostly prompt engineering, RAG, agents, evals, and integration with existing software.
Stack emphasis: Stages 8–14 heavily; basic understanding of 3–7.
Day-to-day: build a Slack bot, integrate Claude into a CRM, ship a doc-search feature, run evals, debug prompts, monitor costs.
Skills:
- Strong software engineering.
- Prompt engineering, RAG, agent design.
- Evaluation and observability.
- Product sensibility.
Companies: every AI-using startup; every traditional company adding AI.
Compensation (2026 range): comparable to senior software engineers; +20–50% in AI-native companies.
2. AI Platform / Infrastructure Engineer
What they do: build the layer that other AI engineers stand on — model serving, fine-tuning pipelines, eval infrastructure, retrieval platforms.
Stack emphasis: Stage 13 (production) and Stage 10 (fine-tuning), plus deep distributed systems.
Day-to-day: scale vLLM clusters, build internal RAG platforms, design fine-tuning pipelines, optimize inference cost, GPU operations.
Skills:
- Distributed systems.
- GPU and CUDA fluency.
- Inference optimization.
- Reliability engineering.
Companies: model providers (Anthropic, OpenAI, Mistral), serving startups (Together, Fireworks), big-tech AI infra teams.
3. ML Research Engineer
What they do: train models, run experiments, push the frontier. The line between “research” and “engineering” is fuzzy at frontier labs.
Stack emphasis: Stages 1–7 deeply, plus research-oriented Stage 10 / 12.
Day-to-day: design training runs, debug loss curves, ablate components, run experiments, write papers, scale training.
Skills:
- Strong ML fundamentals.
- PyTorch / JAX fluency.
- Distributed training expertise.
- Scientific rigor.
Companies: frontier labs (Anthropic, OpenAI, Google DeepMind, Meta FAIR, etc.), top research orgs.
Compensation: very high; competitive with quant trading.
4. Applied Research Scientist
What they do: bridge research and product. Take new techniques, adapt them to specific domains/products, evaluate, productize.
Stack emphasis: Stages 5–13 with research depth.
Day-to-day: read papers, run experiments on internal data, fine-tune models, design specialized retrieval, set up evals.
Skills:
- Research literacy.
- Experimentation discipline.
- Software engineering.
- Communication (across research / product).
Companies: applied AI teams at large companies, well-funded AI startups.
5. AI Product Manager
What they do: not engineers per se, but increasingly technical. Define what to build; coordinate; prioritize.
Stack emphasis: Stages 8–14 conceptually; less code.
Day-to-day: understand user needs, scope features, write specs, coordinate with engineering and design, prioritize roadmap.
Skills:
- User empathy.
- Product strategy.
- Technical literacy.
- Cross-functional communication.
6. AI Safety / Alignment
What they do: research and engineering focused on making AI systems safe — robust, aligned with intended use, free of dangerous capabilities.
Stack emphasis: Stages 7, 10 (RLHF, DPO), 13 (guardrails) deeply; mechanistic interpretability tooling.
Day-to-day: red-team models, develop alignment techniques, evaluate harm, build safety classifiers.
Skills:
- ML fundamentals.
- Adversarial mindset.
- Ethical reasoning.
- Strong writing.
Companies: frontier labs (alignment teams), AI safety orgs (ARC, AI Safety Institute, METR).
7. AI Specialist (Domain-vertical)
What they do: AI applied to a vertical — legal, medical, finance, engineering simulation, science.
Stack emphasis: domain knowledge + Stage 8–14.
Day-to-day: combine domain expertise with AI techniques. Build legal contract review, biology research tools, etc.
Skills:
- Deep domain knowledge.
- AI engineering basics.
- Translation between users and engineers.
Picking a track
A simple frame: which 3 questions do you find yourself asking most?
- “What does the user need? How do we ship?” → Application/Product.
- “How do we run this at scale, cheap and reliable?” → Platform.
- “Can we build a model that does X better?” → Research.
- “How do we make this safe?” → Safety.
- “How do we apply AI to [my domain]?” → Vertical specialist.
Most people end up doing some combination. Early in your career, optimize for what you find energizing.
Skills that matter across all tracks
Regardless of role:
- Software engineering basics — code is still code. Tests, version control, debugging, design.
- Communication — written, verbal, visual.
- Curiosity + iteration — the field moves; you must keep learning.
- Product sense — you’re solving real problems.
- Statistical literacy — not deep stats, but enough to read evals critically.
Salary realities (early 2026)
Wide variance. Rough ranges (US, total comp including stock):
- Junior application engineer: $120k–$180k.
- Senior application engineer at AI-native startup: $200k–$400k.
- Frontier-lab research engineer: $400k–$1M+.
- Top alignment researcher: $400k–$1M+.
Startups offer equity that can dwarf base. Frontier labs offer high cash + stock.
These will move; check your local market.
Career capital in 2026
What compounds:
- Public work: blog posts, open-source contributions, models on HuggingFace, GitHub repos.
- Specific deep skills: e.g. “the person who built the eval system at X.”
- Connections: AI Twitter/Bluesky, conferences, communities.
- Frontier-adjacent experience: if you’ve shipped or trained at scale, opportunities open.
What doesn’t compound as much:
- Generic “I used ChatGPT” credentials.
- Certificates from short bootcamps.
- Conference attendance without contributions.
Career paths that exist
Some are surprising:
- Solo AI engineer building $1M ARR products — increasingly common.
- Independent contractor / fractional AI lead — high day rates for proven specialists.
- Open-source maintainer — community + sponsorship.
- AI educator / creator — content, courses.
- Research scientist with no PhD — rarer but possible at applied teams.
The field is too new and growing too fast for one career model.
Exercise
Pick one role from above. Write down:
- What’s appealing about it?
- What skill gap stands between you and that role?
- What’s a 90-day project that closes that gap and gives you something to point to?
That project is your next move.