But Andrew, what a pleasure to have you here in New Delhi with us at the NDTV World Summit. Thank you for joining us.
No, thank you. I'm really happy to be here in India today.
You know, you need no introduction in terms of how much you represent all of us — being that honest hope in the world of AI. When everyone's painting this image and conjuring this image, Andrew, of doom and gloom in terms of AI, the job apocalypse as some people are calling it. I know, and now everyone in this hall soon will know, that there is a counter theory. Andrew, you're not a believer in the job apocalypse.
Yeah, so I want to speak directly to anyone worried about their job prospects — the student worried if there will be a job for you in the future when you graduate. I want to tell you there will be hope, and it's not nearly as bad as the hype says it will be. A number of businesses have hyped up AI job loss or the AI job apocalypse to try to make themselves look more powerful or to gain PR attention, but the data that's coming out is that AI seems to be leading to more jobs created than destroyed. And so if you keep on learning new skills, the jobs will be there. Frankly, I wish AI worked better, but it just doesn't work that well yet, and it's just not good enough to destroy the vast majority of your jobs.
But is that a concern then, that a lot of big tech companies now, let's say big AI companies, are sort of using this narrative to fuel this whole fear about AI? So when the larger public consumes this on social media and all of this doom and gloom theory about how AGI and ESI are going to come in, and probably systems that we have delegated to will take over one day — a terminator sort of a scenario. That is not very likely. In fact, your theory is that it will augment the existing workforce. It might add jobs in certain sectors. We've already seen that.
You know, of all the different job sectors, the one that AI is affecting the most is software engineering, because AI is really good at writing code. And guess what? Certainly in the United States, the number of software engineers is up, because when we make software engineers much more productive, we want to hire a lot more of them. So as AI starts to enter other job sectors, I think we'll see a similar effect as well — in most, maybe not all, but in many different types of jobs. But the challenge, and there is a big challenge, is that people do need new skills to work in this new AI-enabled economy. And so having ways to help people gain the new skills they need, so we're readying them not for the jobs of 2026, or even heaven forbid the jobs of 2024, but for the jobs of 2028 and beyond. So that's the challenge ahead of us.
That's a key point, Andrew, because everyone's going to Coursera and a lot of other platforms where they're trying to upskill when it comes to AI. Newsrooms are trying to transform themselves with AI transformation courses. Young influencers are teaching us what to do and what not to do in the world of AI. So you think that we are prepping for a 2026 scenario with 2024 skills, probably, but we should be looking at 2028 and beyond to make sure that our workforce here in India — a lot of software engineers that you spoke of — are genuinely armed and equipped to take on whatever's coming their way.
Yeah. So frankly, we have a lot of work to do there around the world. I'm worried about two things. First, even though we hope AI will help learning in the future, I'm going to say something that sounds controversial but is well supported by a growing body of evidence, which is that AI chatbots are terrible for learning. More and more data is coming out that when students use AI, their homework scores go up — no surprise, AI does the work for you — but the final closed-book exam scores go down. One recent study from the University of Pennsylvania's Wharton showed 17% lower grades by students that use AI. And if you take this 17% chatbot learning loss and compound it over the lifetime, or over many years of a student in school, I think if we don't change that, we will be on a trajectory to graduate students that know much less. The good news is, though, I think we can build much better things. It's just not the current generation of chatbots. My second worry is that many university systems and educational systems are slow to change. So all around the world I still see universities teaching curricula from a few years ago, and so it's unfortunate that today it's still possible in many places to get a bachelor's degree in computer science without ever having used an AI coding assistant, or ever made a call over the internet — an API call — to a leading frontier AI model. And today I will not hire a software engineer that doesn't know how to use an AI coding agent. But changing the educational system is hard. But we do need to change it, though, to make sure that we help our students ready themselves for the jobs of today and the future.
You know, when the ChatGPT moment happened in the world of AI, they called it actually the iPhone moment, because everyone thought AI is all about speaking to a chatbot. We were told then — Jensen and top AI leaders around the world said that this is a pyramid and it will build, and the next piece will be on agentic. 2026 was the year of agentic. Now everyone's caught up with agentic already. You have Muse, you have ChatGPT Dots, you have Instinct, which I use all the time. Andrew, you're going to tell us what you're using as well.
Did you know — this is not widely known — but I'm actually the person that coined the term agentic AI. So it was around early 2024 when I thought that this wave of AI that doesn't just blurt out an answer, but takes multiple steps in order to reach a conclusion. And I remember searching on Google for the word 'agentic' and the term was not used. I thought, you know what, I think we should call this agentic AI. And it turns out my team actually said, 'Andrew, the AI world does not need another term.' But I felt this is different. So I kind of ran a secret campaign to get people to call this agentic AI. I started to write in a newsletter and began to talk about why we should call it agentic. And it was controversial back then. But what I was not expecting was that shortly after I was pushing for the use of this term, a number of marketers would get a hold of this term and slap it like a sticker on almost everything in sight. And that caused the hype to take off very rapidly. But the good news is the business applications of agentic AI, while not as fast as the hype, are also taking off rapidly. So I feel pretty good about the trajectory that the world is on.
Okay, so since we haven't scripted any of this and Andrew's had a busy schedule as well, we barely caught up over lunch. But since you've played an instrumental role in the coining of the term agentic, or AI agents and the way we know them, let's have a simple exercise over here — a quick show of hands. Everyone who knows about AI agents, have you guys heard of Muse and Dots, and do you use AI agents? Okay, all right, so then...
But to be transparent, the term AI agents has been around for a long time. I called the new thing agentic AI; it has its roots in agency as well.
But, but, but — so everyone that isn't accustomed or acclimatized to the term agentic, Andrew, we're going to do a simple ChatGPT exercise, which is one where we will explain agentic to this audience like we are explaining it to a seven-year-old, and then we'll explain it to them like we're explaining it to a CEO of a top manufacturing company who wants to transform his company using artificial intelligence. Let's go with the first one. How do I explain it to a seven-year-old?
You know, maybe before the rise of agentic AI, AI used to be something that you prompt and immediately get an answer back from. But with modern AI applications, we don't want AI to just take your prompt and blurt out an answer. We want it to maybe do some web research, maybe look up some data, maybe think for a while, and even reflect and edit itself, and do a sequence of steps before giving you a vastly superior answer. So this type of technology is instrumental to a large number of businesses' use of AI today. But maybe good news and bad news, and why I think there will be no AI job apocalypse, is that it's frankly a lot of work to look through your business and identify where agentic AI workflows can help, and then to get the skilled engineers to go and implement them. Actually at AI Inspire, my business partner Kirsty Tan and I, we work with many businesses that are implementing this very successfully, getting business ROI, but it is hard work. And I think that even though there's hype that we have AGI in a few years and maybe AI will do it all, the reality is I think even a decade from now we'll still be identifying important business process workflows and working hard to implement them.
And if we had to explain this to a CEO, telling him he needs to hire the both of us to do the AI transformation for him?
