4 AI development skills you need, according to Andrew

What happened

Is it time to think outside the box for software engineering AI skills? One of the planet's most noted AI experts recently pinpointed the four most essential skill areas for those seeking careers in AI development. However, some industry observers say the recommendations don't go far enough, suggesting AI skills extend beyond engineering into other business-focused areas.

" "The four skill areas Ng outlined are important, but are not enough on their own," agreed Deepika Sidana, senior manager of software engineering at American Express and professional development director for NYSCD at the Society of Women Engineers. "Building an AI application is only one part of the challenge. " Strong orchestration skills are particularly critical to moving AI into enterprise environments.

Such skills encompass "coordinating models, tools, data, evaluations, observability, human approvals, and fallback paths," said Sidana. " Along with orchestration, key skills include "the ability to harness engineering and deterministic governance; multi-agent orchestration; agent arbitration; AI FinOps; runtime economics; agentic observability; agentic security; and socio-technical systems integration," said Thurai. " The most valuable AI-era engineering skills "increasingly sit outside traditional coding," said Naman Ahuja, software engineer at Meta.

Why it matters

Andrew Ng, founder of Coursera and lecturer at Stanford, formulated a list of essential skills for AI development based on his analysis of more than 10,000 job postings and interviews with AI experts, hiring managers, and recruiters. The list included the following abilities: With the rise of generative, and now agentic, AI, the process of building software has changed drastically, said Ng.

But acquiring the right skills can be a confusing process in the hype-filled environment that has developed around AI. Still, "all developers -- full-stack engineers, data engineers, DevOps engineers, machine learning engineers, and, yes, AI engineers -- will need AI engineering skills," he explained. However, Ng received pushback that his skills-development recommendations needed to consider the bigger picture when introducing AI for problem-solving abilities.

One response to his post puts it this way: "Yikes -- this is way too internally looking. It fails to address the business problem. " While helpful from strictly a development perspective, Ng's recommendations constitute "a dangerously narrow framework for the enterprise, suffering from a massive blind spot of builder bias," said Andy Thurai, founder and AI advisor at The Field CTO.

"Ng's taxonomy is entirely focused on 'Day 1' innovation -- writing the code and getting the model to work. But in enterprise environments, the hardest part of AI is no longer building the intelligence; it is orchestrating, observing, and paying for it.

What to watch

"AI can generate implementation quickly; the harder work is framing the right problem, understanding business constraints, decomposing systems, orchestrating AI and software components, and judging whether an output is actually useful in production". In his work at Meta, "infrastructure decisions often involve balancing reliability, compute efficiency, cost, and user impact, not simply writing technically correct code," said Ahuja. ai. At the same time, "not all rework is waste," he continued.

"When generation is cheap, and the information gained is valuable, prototyping is extremely effective. It takes a holistic view to know which decisions are one-way doors, and to set the criteria under which we'd throw a prototype away. Too often AI-enabled engineers build production-grade software inside a bubble. "