Three years ago, generative AI was a research curiosity. Today it is the most economically significant technology since the smartphone — and the generative AI engineering skills that underpin it are among the scarcest, most valuable, and fastest-growing in the global job market.

I joined OpenAI four years ago and have spent the last two years building and scaling the Applied team — the engineers who take OpenAI's models and turn them into production systems used by millions of people. In that time I have interviewed more than six hundred candidates for generative AI engineering roles. I have watched the candidate pool shift from a handful of specialists who had been researching language models for years, to a much broader community of software engineers, data scientists, and career-switchers who have built genuine, deployable generative AI skills through structured learning and project work.

The barrier to a generative AI career has never been lower. The demand for people who can build with LLMs has never been higher. This guide is the resource I wish existed when the people I interview were starting out — a clear, honest roadmap from wherever you are now to a career in one of the most consequential technology fields of this decade.

📊
The Demand Signal Is Unmistakable

LinkedIn reported a 700% increase in generative AI job postings between January 2023 and January 2026. McKinsey's 2025 AI State report estimates that generative AI could add $2.6–$4.4 trillion annually to the global economy. The World Economic Forum projects that 40% of working hours across all professions will be touched by generative AI within five years — creating massive demand for people who can build, deploy, and govern these systems.

What Is Generative AI?

Definition

Generative AI refers to artificial intelligence systems that learn to produce new content — text, images, audio, video, code, or structured data — by modelling the statistical patterns of large training datasets. Unlike traditional AI systems that classify inputs or predict numerical outputs within a fixed schema, generative models sample from a learned distribution to create novel outputs.

The most commercially significant generative AI systems in 2026 are large language models (LLMs): transformer-based neural networks trained on vast text corpora that have developed an emergent ability to generate fluent, contextually appropriate text across virtually any domain. GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro, and Llama 3 are the dominant models. These systems can write, reason, code, translate, summarise, and converse at a level that was science fiction five years ago.

Evolution

Generative AI has a longer history than the current moment suggests. Variational Autoencoders generated simple images in the 2010s. GANs produced photorealistic faces by 2018. The transformer architecture introduced in "Attention Is All You Need" (2017) enabled the scale that made modern LLMs possible. GPT-2's release in 2019 was the first sign that language generation had crossed a qualitative threshold. The inflection point was ChatGPT in November 2022 — not because it introduced new technology, but because it packaged existing technology in a form that 100 million people could use immediately.

Why It Matters for Careers

Generative AI is a general-purpose technology. Unlike previous waves of enterprise software that solved specific narrow problems, LLMs can be applied to almost any task involving language — customer service, legal analysis, code generation, content creation, data analysis, medical documentation, financial modelling. Every industry with a knowledge work function has a generative AI application waiting to be built. The people who can build those applications are in every organisation's hiring plans.

Why Generative AI Is One of the Fastest-Growing Career Fields

Five structural forces are compounding to make generative AI one of the most durable career opportunities of the next decade:

  • Enterprise adoption is accelerating. As of early 2026, 78% of Fortune 500 companies have active generative AI programmes — up from 12% in 2023. The majority are still in pilot or early production phases, meaning demand for engineers who can take those pilots to production is still years from its peak.
  • The supply-demand gap is widening. There are approximately 85,000 job postings requiring LLM and RAG engineering skills in the US and UK right now. The number of experienced generative AI engineers who can fill those roles is estimated at under 30,000. This gap is not closing quickly because the field is evolving faster than formal education can adapt.
  • The tools are maturing into engineering disciplines. Generative AI now has formal patterns (RAG, agentic systems, RLHF, Constitutional AI), evaluation frameworks, MLOps tooling, and production deployment requirements. This maturation creates durable roles that cannot be automated away easily.
  • Every adjacent profession is being transformed. Software engineers become AI-augmented developers. Product managers become AI product managers. Data scientists become generative AI specialists. The skills that make each transformation successful overlap significantly — making this a field where career-switchers have genuine advantages.
  • The compensation is exceptional and rising. Median generative AI engineer compensation in the US now exceeds $185,000 in base salary. With equity and bonus, total compensation at top-tier companies regularly exceeds $300,000.

Current Generative AI Job Market

The generative AI job market in 2026 has stratified into three tiers worth understanding before you chart your learning path:

Tier 1 — Foundation model companies and top-tier AI labs (OpenAI, Anthropic, DeepMind, Meta AI): These roles require deep ML research backgrounds and often published research. The smallest segment by headcount. Total packages of $500,000–$1.5M+ are not unusual at this level.

Tier 2 — AI-native companies and well-funded AI startups (Perplexity, Cohere, Mistral, Hugging Face, plus thousands of Series A-C AI application companies): The sweet spot for most career-switchers. They require demonstrated ability to build with LLMs — RAG systems, agent systems — but not deep ML research experience. Base salaries of $150,000–$220,000 in the US, £85,000–£145,000 in the UK.

Tier 3 — Enterprise and traditional companies adding AI capabilities: The largest segment by headcount and the most accessible entry point. These roles require applied generative AI skills without the same depth of systems expertise. Base salaries of $110,000–$165,000 in the US, £70,000–£110,000 in the UK. More stable, more structured.

Related Reading

Top Generative AI Career Paths

⚙️
Generative AI Engineer
🔥 Hottest Role 2026
What They Do
Design and build production LLM applications — RAG systems, AI agents, chatbots, and pipelines. Evaluate model output quality. Optimise cost, latency, and accuracy. Deploy and monitor LLM systems in production.
Key Skills
Python, LangChain, RAG, vector databases, LLM APIs, FastAPI, Docker, evaluation frameworks.
Background
Software engineering, ML engineering, or data science with LLM upskilling.
Salary Range
US: $150,000–$220,000 · UK: £85,000–£145,000
✍️
Prompt Engineer
⚡ High Demand
What They Do
Design, test, and optimise prompts and prompt pipelines that reliably produce high-quality outputs from LLMs. Manage prompt libraries, run systematic evaluations, and reduce failure rates in production LLM systems.
Key Skills
Prompt design patterns (CoT, few-shot, structured outputs), evaluation methodology, Python scripting, LLM API familiarity.
Background
Technical writing, linguistics, QA engineering, or product. Lower coding barrier than engineering roles.
Salary Range
US: $105,000–$145,000 · UK: £65,000–£100,000
🧠
LLM Engineer
🔥 Hottest Role 2026
What They Do
Fine-tune and adapt LLMs for specific domains. Work on RLHF, instruction tuning, and model alignment. Evaluate model behaviour systematically. Manage model versioning and deployment pipelines.
Key Skills
PyTorch, Hugging Face Transformers, LoRA/QLoRA fine-tuning, RLHF, model evaluation, GPU infrastructure.
Background
Strong ML engineering foundation required. Typically 2+ years with deep learning frameworks.
Salary Range
US: $165,000–$235,000 · UK: £95,000–£155,000
📋
AI Product Manager
⚡ High Demand
What They Do
Define strategy, roadmap, and requirements for AI-powered products. Bridge technical and business teams. Own AI product metrics, evaluation frameworks, and user experience design. Manage stakeholder expectations around AI capabilities and limitations.
Key Skills
Product strategy, LLM literacy, evaluation design, stakeholder communication, data analysis, user research, Agile.
Background
Product management experience plus technical AI literacy.
Salary Range
US: $140,000–$195,000 · UK: £90,000–£150,000
🏛️
AI Solutions Architect
⚡ High Demand
What They Do
Design end-to-end AI system architectures for enterprise clients or internal teams. Evaluate AI vendor offerings. Define integration patterns, data flows, and infrastructure requirements. Ensure systems meet security and compliance requirements.
Key Skills
System design, cloud platforms (AWS/Azure/GCP), LLM APIs, enterprise integration patterns, security, stakeholder management.
Background
Senior software or cloud architecture experience with AI specialisation layered on top.
Salary Range
US: $155,000–$225,000 · UK: £95,000–£160,000
💼
AI Consultant
📈 Growing Fast
What They Do
Advise organisations on generative AI strategy, use-case identification, vendor selection, governance frameworks, and implementation roadmaps. Typically works across multiple clients and industries simultaneously.
Key Skills
AI strategy, business analysis, process design, change management, technical literacy, client communication.
Background
Management consulting or business analysis with AI upskilling. Or technical AI background with client-facing experience.
Salary Range
US: $130,000–$175,000 · UK: £80,000–£130,000 · Freelance: £800–£1,800/day
⚡
AI Automation Specialist
📈 Fastest Growing Role
What They Do
Design and build AI-powered workflow automations that replace or augment manual business processes — using LLM APIs, no-code AI tools, and custom agents. Works across operations, HR, finance, and customer service to identify automation opportunities and implement them rapidly.
Key Skills
Workflow analysis, LLM APIs, no-code AI platforms (Make, Zapier AI, n8n), Python scripting, process documentation, ROI measurement.
Background
Operations, business analysis, or IT with AI tools knowledge. Lower technical barrier than engineering roles — strong entry point for non-coders.
Salary Range
US: $90,000–$140,000 · UK: £60,000–£100,000 · Freelance: £600–£1,400/day

Generative AI Salary Guide 2026

Salary benchmarks below are base salary only, sourced from LinkedIn Salary, Glassdoor, and Levels.fyi data for 2025–2026. Equity and bonus can add 20–60% on top at well-funded companies.

RoleUS Entry-LevelUS SeniorUK Entry-LevelUK Senior
Generative AI Engineer$130,000–$160,000$185,000–$220,000£75,000–£100,000£115,000–£145,000
LLM Engineer$140,000–$175,000$195,000–$235,000£80,000–£110,000£125,000–£155,000
Prompt Engineer$90,000–$115,000$120,000–$145,000£55,000–£75,000£80,000–£100,000
AI Product Manager$120,000–$150,000$165,000–$195,000£75,000–£100,000£120,000–£150,000
AI Solutions Architect$135,000–$165,000$185,000–$225,000£80,000–£110,000£125,000–£160,000
AI Consultant$105,000–$135,000$150,000–$175,000£65,000–£85,000£100,000–£130,000
AI Automation Specialist$80,000–$100,000$115,000–$140,000£50,000–£70,000£75,000–£100,000
💡
Geographic Premiums

San Francisco and New York typically pay 25–40% above the US national figures. London pays 15–25% above UK national figures. Remote roles at US companies accessible to UK-based engineers often pay US salaries — currently one of the most financially advantageous positions in the market.

Skills Required for Generative AI Careers

Technical Skills

🐍
Python
The universal language of AI engineering. You need fluency — the ability to write clean, functional Python for data processing, API integration, and application logic without searching for syntax.
Required for all GenAI engineering roles
🔌
LLM APIs
OpenAI, Anthropic Claude, Google Gemini, and Mistral APIs are the building blocks of most generative AI applications. Understand streaming, function calling, token management, cost optimisation, and error handling.
Required for all GenAI engineering roles
🎯
Prompt Engineering
Systematic design and testing of prompts — chain-of-thought, few-shot examples, structured output formatting, role assignment, constitutional prompting. Not just writing good prompts, but evaluating and improving them methodically.
Required for all GenAI roles including non-engineering
🧠
RAG (Retrieval-Augmented Generation)
The dominant pattern for grounding LLM outputs in real data. Understand document ingestion, chunking strategies, embedding models, vector stores, retrieval algorithms, and re-ranking.
Core skill for GenAI Engineer and LLM Engineer roles
📦
Vector Databases
FAISS (local), Pinecone (managed), Weaviate (open-source), and Chroma (lightweight) are the most common. Understand similarity search, indexing strategies, hybrid search, and when to use each.
Essential for RAG and semantic search applications
🦜
LangChain / LlamaIndex
The two dominant frameworks for building LLM applications. LangChain is broader (agents, chains, tools); LlamaIndex specialises in data-intensive RAG. Most production applications use one or both.
Core skill for GenAI engineering roles
🤖
AI Agents
Systems that plan and execute multi-step tasks using tools, memory, and reasoning loops. Understanding ReAct, tool-calling, agent evaluation, and multi-agent orchestration with CrewAI or AutoGen is increasingly expected.
Advanced — differentiates senior candidates in 2026
🚀
Deployment & MLOps
FastAPI for serving, Docker for containerisation, basic cloud platform knowledge (AWS, Azure, or GCP), logging, monitoring, and cost tracking. Production GenAI engineers need to own their deployments end-to-end.
Required for production GenAI engineering roles

Business & Soft Skills

Technical skills open doors. The ability to communicate clearly, reason about business problems, and design processes that work for real humans determines how far you go once you are inside.

  • Clear technical communication. Explaining what a RAG system does and why it sometimes hallucinates to a non-technical stakeholder is a skill. Writing documentation that a future engineer can use without contacting you is a skill. Presenting evaluation results with appropriate caveats is a skill. All three are tested in generative AI roles.
  • Problem decomposition. The most common failure mode I see in junior generative AI engineers is reaching for a solution before fully understanding the problem. Defining the problem precisely — what is the input, what is the desired output, what counts as success, what are the failure modes — is the most valuable skill in the first third of any project.
  • Process design. Generative AI applications are process automation systems. Understanding the workflow you are automating — who does it, how, in what order, with what inputs and outputs, and where the failure points are — is essential for building something that works in production.
  • Evaluation mindset. Knowing that your LLM application works is different from knowing how well it works and in what conditions it fails. Designing a systematic evaluation framework — test sets, metrics, human review protocols, regression testing — is worth significantly more than shipping and hoping.
Related Reading

Essential Generative AI Tools

💬
ChatGPT
LLM Interface
OpenAI's flagship product. GPT-4o offers text, vision, and voice. Essential for exploring capabilities, testing prompts, and as an API backend for production applications.
🟠
Claude
LLM Interface
Anthropic's model family. Claude 3.5 Sonnet leads for coding, long-document analysis, and instruction-following. The API is widely used in production applications.
💎
Gemini
LLM Interface
Google's model family with the largest context window. Gemini 1.5 Pro (1M+ tokens) enables whole-codebase analysis and large document processing.
🔍
Perplexity
AI Search
AI-native search with real-time web retrieval. Demonstrates RAG concepts in a consumer product. API useful for research-augmented applications.
🦜
LangChain
Framework
The dominant framework for building LLM chains, RAG pipelines, and agent systems. LangSmith for observability, LangGraph for agent state machines. Learn this first.
🦙
LlamaIndex
Framework
Specialises in data-intensive RAG applications. Superior indexing strategies, query engines, and data connectors for enterprise-scale document systems.
🤗
Hugging Face
Model Hub
The open-source model hub. Home to 500,000+ models, the Transformers library, and the Spaces deployment platform. Essential for fine-tuning and open-source LLM work.
📌
Pinecone
Vector DB
The leading managed vector database for production RAG systems. Fully managed, scales seamlessly, integrates directly with LangChain and LlamaIndex.
📊
LangSmith
Observability
LLM application monitoring, tracing, and evaluation from the LangChain team. Track prompts, outputs, latency, and costs in production — non-negotiable for serious applications.

Generative AI Learning Roadmap

1
Phase 1 · Beginner
Python & LLM Fundamentals
Build Python proficiency to the point where you can write data processing scripts, call APIs, and build simple web applications without friction. Learn how LLMs work conceptually — transformers, tokens, attention, context windows — without needing to implement them. Get comfortable with the OpenAI and Anthropic APIs: sending prompts, handling streaming responses, managing token costs, and working with structured outputs. Build your first simple LLM application — a chatbot, a summarisation tool, or a Q&A system.
PythonOpenAI APIAnthropic APIPrompt EngineeringStreamlitFastAPI basics
Timeline: 6–10 weeks · Outcome: Build and deploy a simple LLM application
2
Phase 2 · Intermediate
RAG Systems & LangChain
Learn LangChain and build your first RAG system from scratch — document ingestion, embedding, vector store indexing, retrieval, and generation. Understand the difference between dense and sparse retrieval. Learn to evaluate RAG output quality using metrics like faithfulness, answer relevancy, and context precision. Build a domain-specific chatbot that retrieves answers from a real document set. Deploy on a cloud platform with proper monitoring via LangSmith.
LangChainLlamaIndexFAISS / PineconeEmbedding modelsRAG evaluationDockerLangSmith
Timeline: 8–12 weeks · Outcome: A production-ready RAG application with an evaluation suite
3
Phase 3 · Advanced
Agents, Fine-Tuning & Production Engineering
Build your first autonomous AI agent — a system that plans, uses tools (web search, code execution, APIs), and executes multi-step tasks. Learn multi-agent orchestration with CrewAI or AutoGen. Add fine-tuning with LoRA or QLoRA for open-source models. Build a complete production system with authentication, rate limiting, cost monitoring, error handling, and automated regression testing. Apply for roles during this phase.
AI AgentsCrewAI / AutoGenLoRA fine-tuningHugging FaceAWS / AzureMLOps
Timeline: 10–14 weeks · Outcome: Autonomous agent + fine-tuned model + production deployment = interview-ready portfolio
💡
The Fastest Path for Software Developers

If you already write Python and have built web applications, skip Phase 1 theory and go directly to LLM API integration. Build a simple Streamlit chatbot in week one. Your software engineering skills mean you can focus immediately on the LLM-specific knowledge — prompting, RAG, evaluation — without the Python foundation work. Most experienced developers can reach Phase 2 outcomes in 8–10 weeks of focused learning.

Projects to Build at Every Level

Beginner Projects

💬
Domain-Specific Chatbot
Build a chatbot for a specific domain — customer service for a fictional e-commerce company, an FAQ bot for a university, or a cooking assistant. Focus on prompt design, conversation history management, and a clean Streamlit interface. Deploy on Streamlit Community Cloud.
OpenAI APIStreamlitPrompt Design
📝
Document Summarisation Tool
Build a tool that accepts PDF or text files and produces structured summaries — key points, action items, and an executive brief. Add a comparison of summary quality across different prompting strategies. Deploy as a Gradio or Streamlit app.
LLM APIspdfplumberPrompt Engineering
🌐
AI Content Generator
Build a tool that generates social media posts, blog outlines, or email campaigns from a brief. Add tone/style controls. Include an LLM-as-judge quality evaluation step — this introduces a critical production skill that most beginners skip.
Anthropic APILLM-as-judge evalStreamlit

Intermediate Projects

🔭
RAG-Based Knowledge Assistant
Ingest a real document corpus into a vector store. Build a retrieval system with hybrid search and re-ranking. Add citation tracking so every answer references the source document. Evaluate with a ragas test suite. This is the single most portfolio-valuable intermediate project in 2026.
LangChainPineconeragas evalFastAPIDocker
📊
AI-Powered Analytics Assistant
Build a natural language interface for data analysis — users describe what they want to know and the system translates that to SQL, executes it, and presents results with AI-generated interpretation. Demonstrates LLM + structured data integration, a common enterprise pattern.
Text-to-SQLPostgreSQLLangChainStreamlit

Advanced Projects

🤖
Autonomous Research Agent
Build an agent that accepts a research question, searches the web and academic databases, reads sources, evaluates relevance, and synthesises a structured report with citations and confidence levels. Evaluate on a test set of 20 research questions. Publish success rate and failure analysis.
LangChain AgentsTavily APIAgent evalFastAPI
🏢
Enterprise GenAI Application (Production-Ready)
Build a complete production-quality generative AI application with authentication, rate limiting, usage analytics, cost controls, a CI/CD pipeline, automated testing, and observability via LangSmith. Architecture diagram, evaluation suite, and deployment documentation required. The portfolio centrepiece for senior roles.
Full-stackAuth0 / JWTDockerAWS / AzureLangSmithCI/CD
Project Depth

Certifications and Learning Resources

DeepLearning.AI — Generative AI with LLMs
Coursera · DeepLearning.AI
Technically rigorous coverage of LLM architecture, fine-tuning (LoRA, PEFT), RLHF, and deployment. Co-developed with AWS. The best technical foundation course for LLM engineering.
Intermediate
AWS Certified Machine Learning Specialty
Amazon Web Services
Covers SageMaker, AWS AI services, ML workflows, and deployment on AWS. Strong for cloud-focused AI engineering. Pairs well with generative AI project work on AWS Bedrock.
Intermediate
Microsoft Azure AI Engineer Associate (AI-102)
Microsoft
Covers Azure OpenAI Service, Cognitive Services, and AI solution design. The most relevant certification for UK and European enterprise AI roles — Azure dominates UK enterprise.
Intermediate
Google Cloud Professional ML Engineer
Google Cloud
Covers Vertex AI, BigQuery ML, GenAI on GCP, and ML pipeline design. Strong for GCP-heavy enterprise environments and roles at Google Cloud partners.
Advanced
Hugging Face NLP Course
Hugging Face (Free)
The best free resource for transformer and LLM engineering. Covers the Transformers library, fine-tuning, tokenizers, and Hugging Face Spaces. Directly applicable to production LLM work.
Beginner–Intermediate · Free
LangChain for LLM Applications
DeepLearning.AI (Free)
Short, practical course on LangChain fundamentals — chains, memory, agents, and retrieval. Free, takes 2–3 hours, and immediately applicable. Start here before longer courses.
Beginner · Free

Common Mistakes Beginners Make

  • ❌
    Learning theory before building anything
    FIX
    Most people spend their first month reading about transformers before writing a single line of LLM code. This is backwards for career purposes. Call the OpenAI API in week one. Build a chatbot in week two. Theory becomes meaningful when you have a concrete system to attach it to.
  • ❌
    Using only one LLM API
    FIX
    Exclusively building on OpenAI's API creates fragility and limits hirability. Build with at least two providers — OpenAI and Anthropic Claude are the minimum. Understand differences in capability, pricing, rate limits, and latency. Production applications should have fallback logic across providers.
  • ❌
    Building without a systematic evaluation approach
    FIX
    "I tested it and it works" is not evaluation. Build a test set of 20–50 representative inputs with expected outputs. Run your system against them. Track performance over time as you change the system. Publish the results in your README. This is the most common gap in intermediate generative AI portfolios.
  • ❌
    Ignoring prompt version control
    FIX
    Most beginners treat prompts as temporary strings, changed whenever something does not work, with no record of what changed or the impact. Production generative AI requires prompt versioning — storing every version, tracking which produced which output, and being able to roll back. LangSmith makes this straightforward.
  • ❌
    Not considering cost and latency from the start
    FIX
    A RAG system costing $0.04 per query seems cheap until it processes 100,000 queries per day. Production generative AI requires cost awareness from day one — track tokens per request, implement caching for repeated queries, use smaller models where the task allows, and monitor cost trends.

How to Build a Generative AI Portfolio

A generative AI portfolio is evaluated on three things: can you build a deployed LLM application?, do you have a systematic approach to evaluation?, and do you understand the production concerns of generative AI — cost, latency, hallucination handling, and observability?

The minimum viable generative AI portfolio for an engineering role is: one RAG-based application, fully deployed with a live demo URL, with a documented evaluation suite, and a clear README explaining the architecture, the retrieval strategy, the evaluation methodology, and the known limitations. That one project, done thoroughly, will advance more applications than five shallow chatbots.

The portfolio structure I recommend: start with one beginner project to establish that you can build and deploy a basic LLM application. Then build your centrepiece RAG project — this is where most of your effort should go. Finally, add either an agent project or a fine-tuning project to demonstrate advanced depth. Document all three thoroughly on GitHub with live demo links. Write a LinkedIn post about each one. Apply during the process of building the third project — do not wait until you feel finished.

The detail that gets candidates a second-round interview: a ragas evaluation suite, a test set of 50 representative questions, and a results table showing faithfulness and answer relevancy scores. This signals engineering discipline that most candidates do not demonstrate.

Future of Generative AI Careers

2026–27
Agentic AI Dominates
Multi-agent systems replace single-LLM workflows in enterprise. Agent engineering becomes a distinct and highly compensated specialisation. Demand for engineers who can build, evaluate, and debug autonomous systems spikes.
2027–28
Multimodal at Scale
Text + image + audio + video models become the standard interface. Engineers who can build multimodal applications command significant premium over text-only specialists.
2028–30
AI Governance & Safety
Regulatory requirements in the EU, UK, and US drive demand for AI Safety Engineers and AI Governance specialists. Technical generative AI experience plus policy literacy becomes extremely valuable.

The most durable career bet in generative AI is not on a specific tool or model — both will be outdated in eighteen months. The durable bet is on the engineering discipline: taking a problem, designing an appropriate LLM-based solution, evaluating it rigorously, deploying it responsibly, and explaining every decision you made. That skill set transfers across whatever the next generation of models looks like.

For more context on where AI careers are heading, see our Future of AI Careers: Top Jobs & Skills Through 2030 and the AI Engineer Career Roadmap.

How Atlia Learning Builds Generative AI Careers

Atlia's Generative AI program covers the complete path from Python fundamentals to production-grade LLM engineering — RAG systems, AI agents, fine-tuning, and cloud deployment. Every project is reviewed by mentors actively building generative AI systems at OpenAI, Anthropic, Google, and AWS.

Students graduate with three deployed, portfolio-ready generative AI projects, a thoroughly optimised GitHub profile, a tailored resume, and career support including mock technical interviews and direct connections to hiring managers in our employer network.

View Generative AI Program Book Free Counselling

PCP: 9 months · $6,000  |  PGP: 12 months · $9,999 · US & UK cohorts

Leila Ahmadi
Director, Generative AI Applied · OpenAI
Leila Ahmadi is Director of Generative AI Applied at OpenAI, where she leads the team responsible for deploying OpenAI models in enterprise and consumer applications. She joined OpenAI four years ago and has scaled the Applied engineering team from eight to over one hundred and forty engineers, hiring across the full spectrum of generative AI roles. Before OpenAI, Leila was a research scientist at Microsoft Research Cambridge, where she worked on large-scale language modelling and dialogue systems. She holds an MSc in Computer Science from Stanford University and a BSc in Mathematics from the University of Cambridge.

Frequently Asked Questions

  • A Generative AI Engineer designs, builds, and deploys production LLM applications — RAG systems, AI agents, chatbots, and pipelines. Core responsibilities include evaluating output quality, optimising cost and latency, and deploying applications with proper monitoring. In 2026 median salaries are $165,000–$220,000 in the US and £95,000–£145,000 in the UK.
  • Starting from Python knowledge: 6–9 months to entry-level. Software developer with API experience: 3–5 months. ML engineer adding LLM skills: 1–3 months. Complete beginner: 9–14 months including Python. These timelines assume 15–20 hours per week of focused practice and project building.
  • US base salaries: Generative AI Engineer $150,000–$220,000; LLM Engineer $165,000–$235,000; AI PM $140,000–$195,000; AI Solutions Architect $155,000–$225,000. UK: GenAI Engineer £85,000–£145,000; LLM Engineer £80,000–£155,000; AI PM £90,000–£150,000. London pays 15–25% above UK national averages. Equity and bonus add 20–60%.
  • No. Employers look for: demonstrated ability to build and deploy LLM applications (portfolio projects), Python proficiency, LLM API experience, LangChain or LlamaIndex familiarity, and clear communication. Career-switchers from software development, data science, and product management are well-positioned with 6–9 months of targeted learning.
  • Traditional AI is discriminative — it classifies from a fixed set of outputs (spam/not spam, churn probability). Generative AI is generative — it produces new content: text, images, code, audio. Discriminative AI learns boundaries between categories; generative AI learns the full probability distribution of data, enabling novel outputs in open-ended natural language.
  • The most employer-recognised: DeepLearning.AI Generative AI with LLMs (technically rigorous), AWS Certified Machine Learning Specialty, Microsoft Azure AI Engineer Associate (AI-102, especially for UK/European roles), Google Cloud Professional ML Engineer, and the free Hugging Face NLP Course. Certifications matter most when paired with portfolio projects — a certification without demonstrated project work carries limited weight.

Conclusion

Generative AI is not a specialisation at the edge of the technology industry. It is becoming the core of how knowledge work gets done across every industry on the planet. The careers it creates are not niche roles for researchers — they are engineering, product, consulting, and strategy roles that exist at every size of company in every sector.

The path into this field is more accessible than most people assume. You do not need a PhD. You do not need to understand transformer mathematics at a deep level. You need Python, you need to build real applications with real LLM APIs, and you need to develop the discipline to evaluate what you build honestly and communicate it clearly.

The three-phase roadmap in this article — fundamentals and API integration, RAG systems and evaluation, agents and production deployment — represents the path that the most successful career-starters I have seen take. It ends with deployed, demonstrable work that proves you can do the job before you have been hired to do it.

Start building. The field will have changed by the time you are done — but the engineering discipline you develop building your first few projects will transfer perfectly to whatever the next generation of tools and models looks like. That is the part that does not expire.