On August 14, Andrew Ng, one of the world’s leading artificial intelligence (AI) pioneers, shared an article titled “The AI Engineering Skills Map” on his X account. This map highlights the AI Engineering skills that he and the team at DeepLearning.AI consider most critical both now and in the near future.

It begins with the observation: “AI allows software to be built in ways very different from 2022, and those who master the skills to leverage this shift will find abundant project and employment opportunities.” To determine which skills are truly worth learning, Ng’s team analyzed more than 10,000 job postings, conducted dozens of structured interviews with AI experts, hiring managers, and recruiters, and incorporated surveys alongside other online data sources. The result, akin to a clustering analysis across a vast dataset, revealed four core skill groups:

  • Building and deploying AI applications: The single biggest difference compared to traditional software is that outputs are difficult to predict. When prompting an LLM or training a model, you cannot know the exact result beforehand. Those who excel in this area understand the foundational building blocks of AI (LLMs, context engineering, RAG, agentic workflows) and use statistical techniques to measure and steer systems toward greater predictability, with disciplined evals and error analysis at the core.

  • Software engineering fundamentals: A deep understanding of how software works helps engineers evaluate trade-offs among cost, scalability, reliability, and speed. More importantly, it enables them to “steer” coding agents using the precise technical vocabulary of software development, rather than resorting to vibe coding without knowing what trade-offs the agent is making.

  • Using coding agents: This skill demands a sound mental model of how agents operate, an understanding of their limitations, knowing how much intervention is enough, managing context, and supplying verifiers/evals so that agents can close the loop autonomously.

  • Product shaping: As agents grow increasingly capable, an engineer’s role shifts from knowing how to build to deciding “what to build,” requiring business acumen and a deep understanding of customer goals.

A notable point regarding terminology: Ng opts for “AI Engineering skills” rather than “AI Engineer” because the scope is far broader. Just as every modern developer needs to understand cloud computing even though only a small fraction carry the title “Cloud Engineer,” all developers—from full-stack and data engineers to DevOps and MLEs—now need AI Engineering skills.

Underpinning it all is a continuous learning mindset: AI continues to evolve at a rapid pace, so skills must be regularly updated to match emerging best practices. Ng noted that he plans to explore each skill in depth and release a more detailed map in future posts.

As raw “code writing” is increasingly assumed by coding agents, value shifts toward both ends of the process: defining the problem correctly at the outset, and measuring and steering unpredictable outputs downstream. This shift also reflects how the market is beginning to redefine the evaluation criteria for engineers working with AI.