TL;DR: Start with Python, SQL, data handling, and software engineering. Add machine-learning fundamentals, embeddings, retrieval, evaluation, APIs, observability, and deployment. Do not treat prompting as the whole AI engineering discipline.
The Learning Sequence
Learn Python and Git, then NumPy, pandas, SQL, probability, and model evaluation. After that, study supervised learning, neural-network concepts, embeddings, vector search, RAG, tool use, and production deployment.
Projects That Prove Progress
Build a data-cleaning notebook, a prediction API, a document search system, and a RAG assistant with citations and an evaluation set. The final project should log latency, failures, cost, and answer quality.
Engineering Skills Matter
AI systems still need API design, queues, caching, authentication, testing, databases, and monitoring. Learn to make a system reliable when the model is uncertain, slow, unavailable, or wrong.
How to Use AI Coding Tools
Use them to explain unfamiliar code, generate test cases, and compare approaches. Keep ownership of data boundaries, security decisions, evaluation criteria, and production behavior.
A Practical Six-Month Path
Spend the first month on Python and data, the next two on ML and model evaluation, then two on LLM applications and RAG, and the final month on deployment and reliability. Publish each project with a short technical explanation.