AI engineer roadmap 2026: what to learn, and in what order
Most AI roadmaps list topics. A useful one also says what to learn first, because each skill depends on the one before it. This roadmap is ordered by those dependencies and checked against what 1,000 Forward Deployed Engineer job posts actually ask for.1
What employers ask for
In Bloomberry’s analysis of 1,000 FDE postings, Python appeared in 66%, AI agents in 35%, cloud platforms (AWS) in 32%, LLM experience in 31% and RAG in 12%. More than half the postings describe working directly with customers.1 The roadmap below covers all of these, in an order where each step makes the next one easier.
The roadmap
- Engineering basics. Python written as backend code (functions, types, error handling), Git and GitHub, the command line, HTTP and a small API with FastAPI, and SQL. Everything after this assumes it.
- How LLMs work, and calling them. Tokens, context windows, sampling and embeddings; calling a model over raw HTTP and an SDK; structured outputs and tool calling; prompts treated as tested application code.
- Retrieval (RAG). Chunking, embeddings, hybrid search, reranking, and measuring retrieval quality. See What is RAG?
- Agents. The agent loop by hand first, then a framework such as LangGraph, then tools over MCP and multi-agent patterns. See What is an AI agent?
- Evaluation and reliability. Eval suites that catch regressions, tracing, cost tracking and guardrails. This is what separates a demo from a system a company will trust.
- Deployment and proof. A system running live at a public URL, documented, plus two more projects. These are what you show in interviews.
Mistakes that cost the most time
- Starting with frameworks. A framework hides the loop you most need to understand. Build it by hand once.
- Collecting certificates instead of shipping. Only about 12% of FDE postings are open to people with under two years of experience;1 the projects you can demo are what get an entry-level application read.
- Skipping evaluation. Without it you cannot tell whether a change made things better or worse.
- Learning in isolation. Customer-facing work appears in more than half the postings.1 Practise explaining what you built to someone non-technical.
How Zero to FDE follows this order
Weeks 1 to 4 cover stages 1 to 4 from scratch, weeks 5 to 8 frameworks, MCP, multi-agent systems and evaluation, and weeks 9 to 12 observability, deployment, your portfolio and interview preparation. An optional four-week enterprise track follows. Week 1 is free and complete, and the full roadmap lists every week.
Sources
Zero to FDE is a 90-day path from an AIML degree to the work Forward Deployed Engineers do: backend basics, LLMs, RAG, agents, evaluation and deployment, with a shipped deliverable every week. Week 1 is free and complete, no card needed.
Start Week 1 free