AI Careers Map¶

In one line: AI work splits into roughly a dozen distinct roles across building, researching, running, deciding and governing; this map tells you what each one does day to day, what it needs, and the most realistic route in from where you are now. Last reviewed: 2026-10-21

The short version¶

• "Working in AI" covers very different jobs. An AI engineer builds products on top of models. An ML engineer trains and serves models. A research scientist invents new methods. A governance lead decides what is allowed. They need different skills.
• The fastest-growing group since 2023 has been AI engineers (also called applied AI or LLM engineers): software engineers who build with model APIs, retrieval, agents and evals. It is the most accessible route for people who already code.
• Research roles still mostly expect a PhD or an equivalent public research record. ML engineering expects solid software skills plus ML fundamentals.
• Non-coding routes are real: AI product manager, solutions architect, AI governance and risk, AI enablement and training, and domain experts who become the AI lead for their function.
• Pick a target role, build two or three public projects that look like that job, and learn in that order. Certificates help most for cloud and vendor-facing roles; see AI certifications guide.

The map¶

flowchart TB
    subgraph Build
        AIE[AI engineer]
        MLE[ML engineer]
        FDE[Solutions architect / forward-deployed engineer]
    end
    subgraph Research
        RS[Research scientist]
        RE[Research engineer]
    end
    subgraph Run
        PLAT[ML platform / MLOps engineer]
        DE[Data engineer for AI]
    end
    subgraph Decide
        PM[AI product manager]
        DS[Data scientist / analyst]
    end
    subgraph Govern
        GOV[AI governance and risk]
        SEC[AI security engineer]
        SAFE[AI safety researcher]
    end
    SWE[Software engineer] --> AIE
    SWE --> MLE
    SWE --> PLAT
    DA[Data analyst] --> DS
    DS --> MLE
    PHD[PhD / research background] --> RS
    PHD --> RE
    AIE --> FDE
    SECBG[Security background] --> SEC
    RISK[Legal / risk / compliance] --> GOV
    BIZ[Product or domain expert] --> PM

Role by role¶

AI engineer (applied AI, LLM engineer)¶

• What you do: build features and products on foundation models: prompts, retrieval (RAG), tool use and agents, structured outputs, evaluation, and production operations. Choose models, control cost, and make things reliable.
• Day to day: writing Python or TypeScript; reading traces of bad answers; improving an eval set; shipping.
• Core skills: software engineering, APIs, RAG, agents, evals, prompt engineering, basic cloud.
• Route in: from software engineering, usually. Build three projects: a RAG app with an eval harness, an agent with tools, and an extraction pipeline.
• Roadmap: AI engineer roadmap. Interview prep: AI engineer interview.

ML engineer¶

• What you do: train, fine-tune, evaluate, optimise and serve models, including classical ML (recommendations, forecasting, fraud) and deep learning. Own pipelines from data to deployed model.
• Core skills: Python, PyTorch, ML fundamentals, data pipelines, model serving, MLOps, GPUs.
• Route in: from software engineering (learn ML fundamentals) or data science (learn production engineering).
• Read: what is machine learning, ML engineer path. Interview prep: ML engineer interview.

Research scientist¶

• What you do: invent and test new methods; publish; set research direction. At frontier labs this includes pre-training, post-training (RLHF, RLVR), interpretability and evaluation.
• Core skills: deep maths and ML theory, experimental design, writing, a publication record.
• Route in: usually a PhD, or an outstanding public record (papers, open-source research code) without one.
• Read: researcher path and the reading roadmap.

Research engineer¶

• What you do: make research run: large-scale training infrastructure, experiments, evaluation harnesses, performance work. Often the most in-demand role at labs.
• Core skills: strong engineering, distributed systems, GPU performance, enough ML to collaborate on research.
• Route in: from strong software or infrastructure engineering plus ML projects; reproducing papers publicly is a good signal. See ZeRO and Megatron-LM and FlashAttention for the flavour of work.

ML platform / MLOps engineer¶

• What you do: build the platform others use: training and serving infrastructure, feature stores, model registries, GPU clusters, LLM gateways, observability.
• Core skills: DevOps, Kubernetes, cloud, infrastructure as code, monitoring, cost control.
• Route in: from DevOps, SRE or platform engineering. See LLMOps and observability and LLM serving.

Data engineer for AI¶

• What you do: pipelines that feed models and retrieval: ingestion, cleaning, chunking and embedding, data quality, governance.
• Route in: from data engineering; add vector search and document processing. See embeddings and vector search.

Data scientist¶

• What you do: analyse data, build and evaluate models, run experiments, and increasingly evaluate AI systems and measure their business impact.
• Route in: from analytics, statistics or science. Strong statistics plus Python and SQL.

Solutions architect / forward-deployed engineer / AI consultant¶

• What you do: help customers or internal teams design and adopt AI systems; part engineer, part adviser. Vendors, cloud providers and consultancies hire many.
• Core skills: broad technical knowledge, architecture, communication, cloud. Vendor certifications are most useful here.
• Route in: from engineering, consulting or pre-sales. See AI system design interview.

AI product manager¶

• What you do: decide what to build and how to measure it; understand model capabilities and limits well enough to set realistic scope; own evals as product quality.
• Route in: from product management (learn the technology hands-on) or from engineering. See AI for leaders.

AI security engineer¶

• What you do: threat-model AI systems, red-team them, secure agents and model supply chains.
• Route in: from application or cloud security. See OWASP LLM Top 10 and the prompt injection red team project.

AI governance, risk and compliance¶

• What you do: set policy, run risk assessments, map to regulation (EU AI Act, sector rules) and standards (ISO/IEC 42001, NIST AI RMF), and answer customer and regulator questions.
• Route in: from legal, privacy, risk, audit or compliance; learn enough of the technology to ask good questions. See AI governance program.

AI safety and alignment researcher¶

• What you do: research how to make advanced AI systems behave as intended: alignment, interpretability, evaluations of dangerous capabilities, oversight.
• Route in: research background, or strong engineering plus self-directed research. Read weak-to-strong generalization, sleeper agents, circuit tracing and frontier safety frameworks.

AI enablement, training and "AI lead for X"¶

• What you do: help a team or function adopt AI well: training, playbooks, champions programs, measuring impact. Often a domain expert (sales, marketing, finance, support) who became the go-to person.
• Route in: be excellent at AI in your current job, then teach it. See AI adoption playbook and AI champions program.

Choosing your target¶

Signals that get you hired¶

• Public projects that look like the job. A repository with a README, an eval table, and honest notes on what failed beats a certificate list.
• Writing. A short blog post explaining a project or a paper shows you understand it.
• Shipping at your current job. "I added an AI feature used by 400 people and cut handling time by X" is the strongest story there is.
• Credentials where they matter. Cloud and vendor certifications help most for solutions, consulting and platform roles. See the certifications guide.

Common mistakes¶

• Aiming at "AI" instead of a role. Choose one target; it decides what to learn.
• Collecting courses instead of building. After the basics, projects teach more.
• Skipping software fundamentals. Most AI jobs are engineering jobs first.
• Assuming a PhD is required for everything. It is for most research scientist roles, not for most AI engineering roles.
• Ignoring your existing expertise. Domain knowledge plus AI skills is a strong, less crowded combination.

Next steps¶

• AI engineer roadmap
• Interviews: AI engineer, ML engineer, AI system design
• AI certifications guide
• Start learning: zero to hero path, AI engineer path, ML engineer path

Sources¶

• Swyx (Shawn Wang), "The Rise of the AI Engineer" (2023), on the AI engineer role definition: https://www.latent.space/p/ai-engineer
• U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Computer and Information Research Scientists (education norms for research roles): https://www.bls.gov/ooh/computer-and-information-technology/computer-and-information-research-scientists.htm
• Chip Huyen, AI Engineering (O'Reilly, 2025), on the scope of AI engineering vs ML engineering: https://www.oreilly.com/library/view/ai-engineering/9781098166298/