Andrew Ng: Core Software Skills Still Essential for AI Agents
Andrew Ng outlines essential engineering areas as AI agents handle more code writing.
Andrew Ng posted about how software engineering fundamentals have changed with agentic coding. He shared an AI Engineering Skills map covering five areas: full-stack, data, architecture, secure plus reliable, and scale plus operate. The core point from the post is that the failure mode is not bad code but a plausible application built to the wrong profile. Other engineers replied that users remain harder to please than compilers and that awareness of the full codebase is still needed to avoid poor choices on nontrivial work.
Combined views
633.3K
8 posts, first seen 5d ago
Andrew Ng: Core Software Skills Still Essential for AI Agents
Andrew Ng outlines essential engineering areas as AI agents handle more code writing.
Andrew Ng posted about how software engineering fundamentals have changed with agentic coding. He shared an AI Engineering Skills map covering five areas: full-stack, data, architecture, secure plus reliable, and scale plus operate. The core point from the post is that the failure mode is not bad code but a plausible application built to the wrong profile. Other engineers replied that users remain harder to please than compilers and that awareness of the full codebase is still needed to avoid poor choices on nontrivial work.
Reactions
7 postsAN 
Andrew Ng@AndrewYNgHow have software engineering fundamentals changed with agentic coding? Here is our AI Engineering Skills map for software engineering fundamentals. https://x.com/i/article/2093384274372419585
- likes: 6.5K
- replies: 185
- bookmarks: 10.2K
- reposts: 1.1K
AL Al-Muwahid@unit0r@AndrewYNg This is a great framing. 👏 Agentic coding doesn't make fundamentals less important—it makes understanding them more important. When AI writes more of the code, engineers need stronger judgment about architecture, correctness, testing, and tradeoffs.
NT Neil Ted@t3d551@polydao Ng's 'wrong profile' line is what we keep seeing on Hermes client builds. The drift shows up at week 3 — original spec said 'summarise tickets', week 3 is 'predict CSAT', and the eval set never moved. That's the profile gap, not the code gap.
AS ashe@ashebytes@max_spero_ same. I think about this all the time I was so good at it
TE Thodoris Evangelakos@TEvangelak96752@max_spero_ I miss the days I'd spend thinking about how to make something work
MF Maxence Frenette@maxencefrenette@max_spero_ I think it's even more fun now