Andrew Ng doesn’t sound like someone caught up in the hype machine. When he talks about AI, the emphasis is on process, talent, and the gritty details of building. Speaking on the No Priors podcast, Ng laid out why AI’s future won’t be decided by scale alone.
Take his view on where AI progress comes from. “There is probably a little bit more juice out of the scalability lemon to be squeezed,” he said, but that is hardly the only path forward. Instead, Ng sees “multiple vectors of progress,” from multimodal models to agentic workflows, even down to experiments in how diffusion models might cross over from generating images to generating text.
That phrase, agentic AI, is one he coined. His team initially pushed back. “One of my team members…said, ‘Andrew, the world does not need you to make up another term,’ but I decided to do it anyway.” For Ng, the point was to move beyond endless debates over definitions. “Rather than debating, ‘Is this agent or not?,’ let’s just say the degrees of agency and say it’s all agentic so you spend our time actually building this.”
Talent is still the biggest roadblock
What actually blocks progress, in his eyes, isn’t a missing piece of technology but the lack of disciplined engineering talent. “The single biggest barrier to getting more agentic AI workflows implemented is actually talent,” he explained. The difference between strong and weak teams, he said, comes down to whether they know how to “drive a systematic error analysis process with eval” or whether they’re “trying things in a more random way that just takes a long time.”
On how much of that process AI itself could automate, Ng was blunt: not much, at least not yet. Human engineers, he argued, are still the ones who need to decide if a misrouted invoice is a serious error or if it’s acceptable to bother the CEO for a signature. “That data set is proprietary,” he said. “It’s just not general knowledge on the internet, so figuring that out is still exciting work to do.”
If there’s one place where Ng is impressed, it’s coding agents. He singled out Claude Code as a personal favorite, calling it “highly autonomous in terms of planning out what to do to build the software, building a checklist, going through it one at a time.” By contrast, flashy demos like “go shop for something for me and browse online” don’t yet pass the production test.. “The economic value of coding is just clear and apparent and massive,” he said, which is why resources and talent have flooded there.
Product managers play a pivotal role
In Ng’s own work at AI Fund, tasks that once took six engineers three months can now be done by two friends over a weekend. The result is a new bottleneck: product management. “The speed of coding is accelerating, the cost is falling, and so increasingly the bottleneck is actually product management,” Ng said. With prototypes built in a day, teams can’t afford to wait a week for user feedback. They rely more on instinct, customer empathy, and quick decisions.
Ng doesn’t romanticize that process either. Early in his career, he tried training engineers to become product managers. “I just foolishly made a bunch of really good engineers feel bad for not being good product managers,” he admitted. The real correlate he looks for now is “very high human empathy,” the ability to synthesize signals and form a mental model of the user.
When it comes to founders, Ng believes the technical edge matters more than ever. “So many things that the world used to do in 2022 just do not work in 2025,” he said. Unless a leader “has a good feel for what this technology can and cannot do, it’s really difficult to think about strategy and where to lead the company.”
Work ethic, too, is a trait he’s reluctant but willing to defend. “I even now feel a little bit nervous saying that because in some parts of society it’s considered not politically correct… but the factual reality is people that work hard accomplish a lot more.”
For Ng, the future of work is about small, skilled teams amplified by AI tools. He recounted hiring a fresh graduate over a 10-year veteran simply because the younger candidate had mastered AI tools. “Turned out a great decision,” he said. Yet the very best engineers, he added, are the ones with long experience who are also “really on top of AI tools.”
The broader prediction he offered is: those who embrace AI will become far more powerful than most people expect. “Two years ago, who would have realized that software engineers would be as productive as they are today when they embrace AI?” he asked. Looking ahead, he sees “people of all sorts of job functions…so much more powerful and so much more capable than they probably even imagined.”
