AI can do about a third of your job, and that makes the rest worth more

Economists who broke jobs down into individual tasks found that AI can do roughly 30 to 40% of the work in many jobs. Ng's point is that the remaining part, the part only a person can do, becomes more valuable, not less. The real risk is not AI itself. The risk is working next to people who use AI while you don't.

“That 60% that a human does has become even more valuable.”

Stop doing the part AI can do, and grow into the part it can't

Ng uses software engineers as the early example. AI writes code very well, yet job openings for engineers went up. The ones in trouble are those who still work the way they did before ChatGPT. His advice applies to every profession: hand the repetitive part to AI and spend the saved time on the skills AI does not have.

Building with AI is no longer only for engineers

This is the change Ng calls one of the most important. Making your own tools with AI has become so much easier that marketers, recruiters, HR specialists and operations people should do it too. He believes the people who learn to build will get more done in their role and simply have more fun at work.

Your context is the advantage AI cannot copy

When AI brainstorms, Ng says, you get a couple of good ideas and several bad ones. You can tell which is which because you know things AI does not: how your customers react, what your manager cares about, what failed last year. He sees this context as the real reason AI will not replace people any time soon. Your experience is what turns AI output into good work.

“Humans have a massive context advantage compared to AI.”

Real examples: his marketing and finance teams build their own tools

A marketer on Ng's team built an app that searches the web for related articles before he writes, and a dashboard that flags new trends. One of his CFOs noticed her team spent hours every week opening documents and copying numbers. She built an automation that runs on a schedule, opens the files, checks that the numbers match and alerts the team when something needs attention. Nobody waited for an engineer.

Employers now hire for agency

Ng's team mapped the skills in AI job descriptions and found something he did not expect in the data. More and more employers ask for people with a strong sense of agency, people who spot a problem and build something to fix it without waiting for instructions. When he interviews marketers, he asks what they have built.

Use AI to work faster, but learn on purpose

Ng says something he had not said publicly before. When AI simply does the work for you, you forget the answer a few months later. Students who use AI get better homework scores but remember less. So AI is great for getting things done, and real skills still come from practice, ideally with someone guiding you.

“AI models are terrible for learning.”

The biggest opportunity: learn AI, build fast, talk to customers

Because building has become so cheap, Ng says the hard part is now deciding what to build. He calls it the product management bottleneck. The people who win are the ones who understand a real customer problem, build a solution quickly with AI and keep improving it. That applies to founders, to freelancers and to anyone inside a company.

“Learn AI, build fast, and talk to customers.”