Andrew Ng on AI fear-mongering, jobs, education, and the biggest opportunities in 2026
Marina Mogilko interviews AI pioneer Andrew Ng on her Silicon Valley Girl channel.
Summary
Marina Mogilko interviews Andrew Ng — co-founder of Google Brain, Coursera, and deeplearning.ai — for her Silicon Valley Girl channel. Ng argues that much of the public fear around AI has been deliberately manufactured by a handful of large AI companies seeking regulatory capture to protect their expensive proprietary models from open-source competition. He contends that the "job apocalypse" narrative is false, that AI currently cannot replace the majority of human work, and that humans retain a decisive "context advantage" over AI for the foreseeable future. He also makes the striking claim that AI models, as most people currently use them, are actively harmful to learning and retention — a position he backs with emerging study data. He closes by discussing AGI definitions, children and AI, privacy, and the biggest opportunities for builders in 2026.
Key Takeaways
FULL TRANSCRIPT
The origins of AI fear-mongering
Marina Mogilko: You're one of the voices in AI who comes with a huge background in machine learning and teaching AI, and you're also a positive voice. This is something I've been seeing especially this summer — how negative the conversation has become, especially on social media. When I post about AI, people talk to me about data centers and job loss. Why do you think this wave started recently? What do you think the causes are?
Andrew Ng: There's been a lot of misinformation about AI, and the root cause of a lot of this is an unfortunate attempt that started two or three years ago — I think it was PR and regulatory capture. It turns out that one of the most valuable things in AI right now is the giant large language models that some companies have trained. But if you spend billions of dollars training a model, it's really inconvenient if someone else trains a model and wants to give it to anyone in the world to use for free. So a handful of leading AI companies, I think, have been very loud voices of fear-mongering around AI, trying to get regulations passed to create an unfair playing field that favors incumbents — so that we all have to pay a high toll for the use of AI, while stifling other teams, be it researchers or other companies that want to give away open-source models that anyone could use much more cheaply.
Unfortunately, fear-mongering works. When you go and say AI is like nuclear weapons — an analogy that has no basis in fact; what do they even have to do with each other? — or when you cherry-pick cases of AI making a misstep and make it much bigger than it is, or even spread misinformation about how AI uses data centers and uses far more water than the actual reality, this drumbeat of fear-based messaging has skewed societal perception to be really negative on AI. That's unfortunate because it's slowing down American adoption of AI, making America less competitive. Unless we get the truth about AI out there — which is that it's a fantastic benefit with some problems, but not nearly the problems they're blown up to be — it will hurt individuals.
The job apocalypse myth
Marina: I'm going to read out some of the problems that people are highlighting. Job loss and inequality — what do you think?
Andrew: The job apocalypse — this idea that AI will take over 50% of jobs, people will be out of work, rioting in the streets — that's just not going to happen. With every wave of technology, including AI, the skills we need to do great work shift. AI is changing job professions, but I wish AI worked well enough to justify that fear. A handful of businesses want to hype up AI and say they have superintelligence or artificial general intelligence or whatever, and that it can do all the stuff that humans do. I wish AI worked better — we're just not good enough to make AI do everything a human does.
If you look at the analysis of jobs, economists like my friend Erik Brynjolfsson at Stanford and Andy McAfee at MIT have analyzed many people's jobs, breaking them down into individual tasks. Maybe AI could do 30 to 40% of many jobs. What that means is that the 60% that a human does has become even more valuable, because it's an economic complement to the 30 to 40% that's now cheaper. What will happen is that people who use AI will replace people who don't use AI — but AI is not in a position, for the vast majority of jobs, to replace people outright.
Of all the different professions, the one most affected by AI right now is software engineering, because AI is actually fantastic at writing code. And what we see is that the number of job openings in software engineering is up — contrary to what the doom fear-mongers would say. AI is not actually able to replace software engineers, and all the good software engineers I know are busier than ever. The flip side is that if someone is still writing code like it's 2022, before ChatGPT, they're in trouble. They need new skills. Don't do the stuff that 30 to 40% of AI can automate — let AI do that. But then gain the skills to do the other 60 to 70% that AI cannot do.
Advice for new graduates
Marina: What would your advice be to new graduates? I talked to Erik on this podcast and he mentioned there's not really a lot of impact on the job market except for people from 18 to 25 who just graduated. What would be your advice to those people who don't yet have the expertise to strategize in their job — who can only do the manual work that AI can do as well?
Andrew: One real challenge for fresh college grads is that the university system is slow to adapt. I love academia — I think we should all support universities. But when AI comes and transforms the way software is written, universities often take a year or two for faculty to master the skills, then create new courses, get curriculum committee approval, get the faculty senate to vote. It just takes years. That speed of change in academia is very poorly matched to the speed of change in AI. Sadly, many universities are still teaching students to be ready for the jobs of 2022, when we shouldn't even be teaching them for the jobs of 2026 — we should be teaching them for the jobs of 2028 and beyond.
What this means is that the job openings are there. Tons of employers I know just can't find enough skilled people at any level of seniority. In my office right now, we have a lot of interns — current college students, fresh college grads, and we even had one high school intern — and they're amazing and productive. But the key is they're all very AI-native. They all use AI tools to do the things AI can do, and then lean in to doing the things that humans can do that AI can't, for a long time to come. There's plenty of work for people to do.
My advice to fresh college grads or people currently in college is: by all means, work hard in classes, get good grades, learn from the instructors. But to the extent that there are still additional skills that the university has not yet adapted to teaching, find other ways to learn online — from Coursera, deeplearning.ai, Udemy, or other places where you can gain the more cutting-edge skills, especially AI skills, that universities have not yet worked into their curricula.
Agency, building, and the evolving workplace
Andrew: I want to say one other thing. If you look at the skill map changes, one of the most important shifts is that it's so much easier to build with AI than before. When something becomes much easier, a lot more people should do it. Not only should professional software engineers build software with AI — it's becoming much easier for everyone to build with AI. People who embrace that will be more productive, will accomplish more, and I think will have more fun than those who don't. AI lets you build really fast. For people who are not just software engineers — marketers, recruiters, HR professionals, operations specialists — if they learn to build with AI, they'll just do so much more, whatever their job role is.
Marina: How do you measure the increase in productivity when you deploy AI? Do you have KPIs in your company?
Andrew: I wish there were a simple answer. I find that the business outcome of AI is more a function of the business than of the AI. For some it may be increased customer growth and retention, or faster service to customers, or increased accuracy in some tasks. The KPIs tend to be related to the business rather than to the AI itself.
Marina: That's an interesting point, because we've been deploying AI actively in my company, and for me as a media company it's probably the amount of views or output. It's interesting how different people measure it — even revenue, if you're becoming more effective with how you make money. Actually, how are you using AI in your business?
Andrew: Oh my goodness. First of all, we have Claude for all of us, and we have certain projects for every social media platform we're on. For example, for this podcast, we have a project called Guests, and it knows all the analytics from previous guests. It has certain criteria by which we rank every single person who comes to the podcast — whether they're cited, whether they have a certain opinion on AI, whether they've been active with AI in their company, or if they're a recent AI founder. It gives them different weights and comes up with a grade out of 40. Forty meaning tier one, thirty meaning tier three, and so on. Then we have another one that analyzes every single podcast and gives me tips on how to ask questions.
Marina: Same for Instagram, same for LinkedIn. It has my tone of voice, my personal dossier, my business strategy. So whenever it writes something, it knows all the facts about me and how I sound. Every social media platform is run by a person — a person makes the strategic call. What I'm working on right now is closing the loop, because sometimes I send feedback in a chat on Telegram, and I really want AI to be able to learn continuously from that feedback and just know my taste better.
Andrew: There's one thing I see a lot in AI, which is that as a data science or brainstorming partner, it often comes up with one or two good ideas, two or three mediocre ones, and four atrocious ones. Sometimes you wonder how the AI could have thought that was even a plausible idea. This relates to the job apocalypse point, which is that for a long time, humans — you, me, everyone watching this — will have a significant context advantage over AI. You know something that's incredibly obvious to you — that was an awful idea — but the AI did not. One of the reasons why AI will not replace our jobs or large businesses anytime soon is because humans have a massive context advantage compared to AI. We know so much from our years of experience — we talked to a customer, we saw the funny facial expression that told us they didn't like this, or our manager said they really cared about something. Almost all humans just know a lot of stuff for which the plumbing does not exist, and I don't think will exist for the foreseeable future, for AI to access.
People sometimes talk about the importance of human judgment or human taste, and some wonder what taste even is — whether it's a fuzzy thing. But to me, the technical thing that underlies why humans have better judgment and better taste than AI is this context advantage. Because this is a long-term advantage that no one is going to solve in a few years, this is why we need a lot more humans with that judgment and taste to keep on complementing AI.
Marina: Doesn't this make education even more important? Because education gives us context. Another thing I'm hearing about AI is that you won't need education because all the information is at your fingertips — you just ask ChatGPT. But when you say context and taste, for me that's years of acquiring knowledge and learning from the best and seeing how they perform, versus just asking a chat.
AI is bad for learning
Andrew: I'm going to say something that may be controversial. I don't know if I've said this publicly before, but I think it's true: AI models are terrible for learning. I know people think AI is wonderful at getting things done — use it all the time, love it. But all the data coming out is very clear. Students score higher on homeworks when they use AI — great, higher homework scores. But their long-term retention and performance is much worse, because the AI is doing the work for them. More and more studies are coming out to back this up.
Wikipedia is a wonderful tool with tons of facts. Web search is a wonderful tool with tons of facts. But it turns out that when you ask AI to do work for you, you're cognitively offloading to AI, which is great because that's how society moves forward and gets work done — but human retention is much worse. It's just so clear that LLMs, as they are most commonly used, are terrible for learning. I'm not saying there's no way to use them in a way that is good for learning — I think there are ways. But even for myself, there are so many things I've asked an AI model over the last six months — building some project, how does this front-end backend component work, give me the answer, get the job done, it was fantastic — but six months later I don't remember the answer when I need to redo that component, so I ask AI again. The data is really clear: we should stop thinking of AI as helpful for learning, at least in the vast majority of ways that the vast majority of people are using AI models today. It's absolutely terrible for learning.
Marina: But you're building a company to help solve that, right? The one-to-one tutoring with AI — is that what you just announced with the $100 million investment from Coursera?
Andrew: Yes. I'm excited about leading a new organization called LearnVector, which is focused on building new learning experiences that are much more one-to-one than one-to-many. Fifteen years ago I was privileged to participate in the online courses movement, which I think changed the way a lot of people learn. But that was, and still remains, largely a one-to-many experience where everyone watches the same video — which actually works well. But the technology now exists to create much more personalized, customized, one-to-one experiences, and our team is working hard on that. I think we'll have a lot more to show by early next year.
When I think about human skill development, because AI has so heavily impacted software engineering, what we see happening in the job market for software engineering is a harbinger — a forerunner — of what we'll see in other disciplines as well. In software engineering, people need to learn new skills, but when they do, they are thriving, creating more value, getting raises, and doing even more exciting projects. I'm seeing early signs of this in other disciplines too. For example, most front-end and back-end developers have now become full-stack developers because of AI — they can take on broader scope. I'm seeing early signs of this in marketing, where someone who did marketing coordination is now able to become more of a full-cycle marketer. I'm seeing recruiters become more full-cycle, doing end-to-end recruiting.
The good news and bad news is that for people to step up to these broader roles, they do need to learn AI skills — but also disciplinary skills, like how to do the other parts of marketing, recruiting, software engineering, or AI engineering. This actually creates a heavy need for people to gain new skills. But when they do — both AI skills and disciplinary skills — they can do much more, hopefully have more fun, work on more exciting projects, and hopefully get paid more as well.
One reason I worry about the fear-mongering is that I got an email from someone who was about to enter college, saying he was really struggling with what to major in because in four years, won't AI do all this and everything he learns will be obsolete? The answer is no — of course it won't all be obsolete. But when we keep pushing these fear messages, we make people wonder if they will even be relevant, and it makes people not lean in to gain the skills that would put them in a much better position. These fear-mongering messages are distorting how many people — including high school students, college students, fresh grads — think about the economy. Making people give up is one of the worst things we could be doing in this era, when people who lean in will thrive.
What to study and the importance of agency
Marina: What would you reply to that email? What would you say is the best major to study now to thrive in the AI era? Do you think it's going deep into a niche, or broader computer science so you can acquire AI skills really fast?
Andrew: I don't know what's the best major — there are an awful lot of great majors. It's like asking what's the best job in the world. My daughter wants to be an astronaut. I don't know if she can major in becoming an astronaut — I'll have to think about that. When she gets older she may change her mind. I see so many opportunities across so many job roles. It all seems very exciting to me. But do learn AI, and do learn to build with AI.
The other thing my team has been working on is an AI engineering skills map, trying to map out the most important skills for AI engineering. One thing I felt intuitively, but was surprised to see show up in the data, was that a lot more job descriptions seem to be saying they want people who demonstrate a very high sense of agency. With AI, there are a lot more opportunities for individuals to spot problems and go build something to solve them. We're really evolving — we've long been evolving, but we're accelerating — into a positive era where people don't just sit around and wait for their boss to tell them what to do.
Marina: This is what I've been feeling a lot, especially since we started doing remote work. I want people to be entrepreneurs within their niche. If you're helping me with LinkedIn, you're an entrepreneur there — you can hire more contractors, deploy different tools, make the strategic decision about whether a topic is good or not. I really think we're moving into a job market where everyone is kind of independent in their workplace. Do you agree?
Andrew: I think people will have much more autonomy and creativity, so I agree with that. I'd even go one step further. I talk to a lot of engineers and others in large companies who tell me their manager tells them to stay in their swim lane — they have a creative idea, but the manager says no, focus on this one thing, often because their manager's career depends on it. But the number of opportunities for people to spot things outside the swim lane and then responsibly explore how to get them done — that feels very exciting to me. I think that in the future, businesses that set up a culture encouraging people to learn AI, build fast responsibly, and talk to customers will drive a lot more value than the more hierarchical, siloed organizations.
Marina: It starts with hiring the right people and then nurturing this in your organization. When you say learn how to use AI and become proficient with AI, can you give me some benchmarks? Like for a marketer or knowledge worker who is advanced with AI — what are you looking for when you're interviewing that person?
Andrew: My team is probably ahead of the curve, but all of my marketers know how to code. As part of how I interview marketers, we ask them what they've built. If they have not built any software —
Marina: If it's a dashboard, is that good or bad? Is it too basic?
Andrew: A dashboard — again, my team is probably somewhat ahead of the curve. All of my marketers have built much more specific things. One of our marketing team members was talking about a tool he'd built: when he's considering writing an article on something, it crawls the web, finds related work, and he has a custom desktop app — he actually built a desktop app that runs on his Mac — to highlight related articles for him. Then you can chat with the whole system, navigate what he's writing as well as the related work. He also had a large dashboard for trolling the internet to highlight exciting things that are popping up. Even on my team, that marketer is ahead of the curve.
Marina: That's great to hear. Any other interesting use cases that will inspire people to build something similar?
Andrew: My finance team uses AI extensively. One of my CFOs realized that her team was spending hours every week clicking through documents, opening files, copying and pasting numbers. So she started building automation scripts that run on a routine — they automatically open files, check what's in there, check for consistency, and highlight to her team if there's something they need to be paying attention to, or if a new document has shown up. Rather than waiting around for an engineer to do the work for them, the team's ability to build not just dashboards but data management infrastructure — ingesting data, alerting them if something's happening — has been really valuable.
My recruiting team — we actually have recruiting engineers, which are professional engineers who sit within the recruiting team and are building very sophisticated tools for recruiting. This is another trend: marketers, recruiters, HR professionals, and ops people should all learn AI. But when you take an engineer and embed them in these teams, that further accelerates what you can do.
Marina: We do the same. We start with something basic, build it ourselves, then we hit a wall, an engineer comes in, and we build it further.
Andrew: When you look at not just software engineers but recruiting engineers, marketing engineers, HR engineers — there's so much valuable engineering work that can now be done. I'm just not worried about running out of engineering jobs. All my friends are so busy. We think, how could we ever run out of engineering jobs?
Privacy, data, and local AI models
Marina: You touched on something that is actually one of the fears when we talk about financial information — how much you're giving to AI. I gave Perplexity permission to scan my Fidelity account so it can track my portfolio and tell me when to rebalance. It doesn't do anything on my behalf, but it has access. Do you think there's any problem with that?
Andrew: This is complicated. AI and privacy is a complex area, and it depends a lot on the company you're sharing your data with. I trust all the hyperscalers to really follow their terms of service and do what they say. My personal opinion — not giving legal or business advice — but I'd be shocked if the largest hyperscalers published terms of service with some privacy notice and then breached that, because that would not be the culture and would be so damaging to their long-term business model.
On the other side, there's been at least one company I won't name that seems to occasionally change its terms of service. You go to the website and a pop-up says, "Hey, we changed the terms of service to retain your data or train on your data," and if you aren't paying attention and click the wrong button, they've suddenly given themselves permission to access your data in a way I'm not comfortable with. I handle some sensitive information, so I tend to be very careful with businesses where I don't feel that the culture, the DNA, and frankly the long-term business model is as tied to protecting individual user privacy as the hyperscalers.
I see businesses grappling with this too. For example, one of my teams, AI Fund, works with very large corporations including banks with incredibly sensitive financial data. As you can imagine, we and our clients do not willy-nilly share really sensitive, often material non-public information with frontier labs without very careful thinking about guardrails and privacy.
Marina: So trusting hyperscalers, but also — another thing you can do is download an open-source model and run it on your computer, and then it just stays on your computer, right?
Andrew: Yes. A lot of banks will actually run things in a virtual private cloud or on-premises so it never even leaves their control. For individuals, for really sensitive things, I sometimes run a local model. It's been interesting — with the open-weight models, some of the latest ones are approaching frontier capability and are actually small enough to run locally. They're really good models now.
Marina: The one from Meta, right? The recent one?
Andrew: Yes, Meta's Llama is a good model. Also the latest version of Qwen is very good. But frankly these models change every other week, so the best practice is not to get stuck on one but to keep trying new models.
Marina: So basically, when there's a situation where you don't trust anyone, you run a local model, and this is how you keep your data safe.
Andrew: I do trust the hyperscalers, but sometimes for literally material non-public information that I just can't send to the cloud, I'll either do it manually without AI help, or if I really need to use AI, I'll very carefully use only a local model.
Loss of human control over AI
Marina: What about loss of human control over AI? I talked to Yoshua Bengio, who is very concerned about open, unregulated AI, and he painted some very scary pictures of AI taking over control — the scenario being that we can't control something that's smarter than us, and if AI gets smarter and smarter, where do we end up? What do you think about that?
Andrew: I think about something else that we can't control — airplanes. No one can build an airplane that you can fly perfectly. Winds will buffet it around. In the early days of developing airplanes, some crashed and people died, and it was tragic and awful. But through those early lessons learned, we learned to control airplanes better and better, so that today we can mostly get in an airplane and not fear too much for our lives.
It's really like that with AI too. No one can perfectly control AI because it generates tokens or outputs that are a little bit random — we don't really know exactly what it'll do. But as we run them, there have been a small number of mishaps, which is unfortunate, and some have done real damage. But the way we engineer almost any system — from an airplane to electrical circuits to now AI — is to carefully grow their capabilities so that we have a controlled environment in which to measure what's wrong and then shape it to make sure we can control it well enough that it behaves responsibly and safely. To this day we can't perfectly control any airplane, and we will never perfectly control AI either. But I think we are certainly controlling them well enough that this loss of control doesn't feel like science fiction.
Marina: What about deepfakes?
Andrew: Deepfakes are a problem. One of the most disgusting things I've ever seen or heard of is non-consensual intimate deepfake imagery. I'm really glad that the US Congress has been moving on this. Let's pass laws. Get rid of that. Penalties for that. There are some really problematic uses of AI that we should outlaw and heavily penalize. Let's just get rid of that.
Children and AI
Marina: What do you think about children and social connection when it comes to kids using AI? We've seen with social media how people are doom-scrolling all day. My daughter, who is five years old, whenever I don't have an answer, she says, "Ask ChatGPT." She thinks ChatGPT knows everything. What would you say about kids' future with AI?
Andrew: First, I think kids have a bright future. It's such an exciting time to be a child, to grow up in this environment with tools that none of us ever had before. At the same time, we've seen that social media — which I think has probably been blamed a bit more than it deserves, but does deserve some blame — has not been great for kids. I actually worry a lot about AI damaging learning. I have a five-year-old and a seven-year-old. When I teach them math, they're young enough that I can basically not let them use a calculator — I can say, "How do you multiply these numbers?" and practice that with them. But as they get a little older, I worry a lot about students cognitively offloading to AI in a way that damages long-term learning retention.
At the same time, I actually built an app. I didn't like any of the free online typing-learning tools, so I built my own to have my daughter learn to type. She's actually getting pretty decent now for a seven-year-old — she can type all the lowercase letters. The shift and uppercase letters are a little bit not quite there yet. But I think this unlocks responsible, adult-supervised use of online tools. It's really tricky, though. Adult-supervised use of digital tools seems like a great thing for kids, but too many adults don't have time to supervise the use of the tools, and then the incentives of social media do funny things.
Biggest opportunities in AI for 2026
Marina: You mentioned the fears, and we talked about how you can improve your work with AI. Can you name some of the biggest opportunities in AI in 2026 for people who want to build — for an individual who wants to build?
Andrew: I don't think it's one size fits all, but because the cost of building has plummeted, I encourage people to learn AI, build fast, and talk to customers. I find myself building things every week, every weekend, because I or someone on our team has some problem and I have some idea for building some AI thing to automate it. Last weekend, I was using a frontier model to analyze a lot of our key business metrics because I didn't have time to do it myself — measuring de-trended key business metrics — and I didn't have time to go find a data scientist to work with me on it. So I used a variety of frontier models, being really careful about their data retention policies. I did not use models with data retention policies I don't like, in order to analyze the data.
What's happening with AI is that the cost of building has plummeted, so the challenge is shifting to deciding what to build — which I've been calling the product management bottleneck. Founders, engineers, and product managers who can talk to customers, develop the taste and judgment on what to build, and then build with AI and iterate quickly — I think there's just a ton of exciting things to do.
Marina: You've been starting so many companies. When I looked at your portfolio — do you think for beginners, when you said you built something over the weekend, how do you decide what to focus on? Or can you pursue multiple ideas because of AI now, playing in different companies at the same time?
Andrew: Building a company is still really, really hard, and there's a lot to be said for single-threaded leadership — someone who's fully focused on just one thing. Over a weekend I can often build an LLM wrapper or a simple application, but I wish it was that easy to build a large company. Building something meaningful often takes either real technical depth or deep customer insight and integration with customers. Yes, we can now use AI to code something in a few hours, but that's a small piece of the puzzle. Spending time understanding the technical complexity and building really complex software takes months, maybe years. Having that deep customer insight to decide what to build also just takes a lot of talking to people, reading facial expressions, doing surveys, doing that over and over until you figure out what to build. There's a lot of value in sampling widely, but then having that focus to go really deep in a couple of sectors still seems important for building a business.
What is AGI, and when will we reach it?
Marina: My last question — I know we don't have much time, but I wanted to ask you about AGI, because people use this word so much. Jensen Huang said we've already reached AGI. You've said it's decades away. What's the one criterion by which you'll say we've reached AGI?
Andrew: Different people say we've reached AGI at different times because of different definitions. The definition I'm most familiar with is AI that could do any intellectual task that a human can. The human brain can take five years to study and produce a PhD thesis — can AI write a PhD thesis? A human can learn to drive a truck through a dense rainforest with tens of minutes of practice — when can AI drive in a new environment with tens of minutes of practice? It feels like there's a long list of things that AI cannot do, for what feels to me like decades. I hope it's only decades — it may turn out to be longer. That's why I think, for that definition of AGI, it's still very far away.
It turns out that because of economic incentives, some companies have had an incentive to try to declare reaching AGI earlier. OpenAI had an economic agreement — which has actually been renegotiated now, so that's gone — but they had an economic incentive to try to declare reaching AGI sooner. If you come up with other definitions of AGI, depending on how far you lower the bar, then you could totally have already reached AGI, or even have reached it 30 years ago, depending on how you want to define it.
Marina: True. Andrew, thank you so much for this positive and very applicable conversation. I like when you watch something and then go measure yourself against what people are doing with AI, look at your own process, and maybe expand it. Thank you so much for sharing what your team is doing and for your insights.
Andrew: Given the huge benefits of AI to come, I hope whoever was watching this is motivated to really go learn AI, apply it, and even go build something.