Good afternoon or good evening or good morning wherever you might be listening. Welcome to a new edition of Identity Matters. I'm Mark McClain, CEO and founder at SailPoint, and I'm very pleased to be joined today by Ethan Mollick. Ethan is a professor at the Wharton School of Business, very impressive, and an innovation and AI expert. And I checked and I think he actually invented the internet, not Al Gore, but anyways, we'll talk about that later. Ethan, welcome to Identity Matters. So glad to have you.
Thrilled to be here. Thank you.
So, we're going to talk about something that everybody's talking about, which is AI and the future of work. And I think a couple of directions we might take here. One is, as folks know, we're in the identity security business and there's a lot of discussion about AI and how secure it is, generative AI, but we are a company full of intelligence workers, you know, software developers and highly trained customer service people and sales people, and everybody's wondering what does this mean for me? So, I'd love to kind of poke at both of these as we get into it. Let's start about just the general impact of generative AI particularly. Obviously, AI's been around a while, but generative AI on the workforce. What kinds of things are people asking you about? What are you seeing that companies are looking at? What are the concerns? Let's just talk about what this is going to do to the workforce.
I mean, so obviously a big question. The thing to know about generative AI is it's a general purpose technology, right? So, it's going to affect everything in different kinds of ways and ways that are really hard to predict. I'm deep into this as an expert. I talk to the AI labs all the time. I talk to policy makers and the answer is nobody knows anything. So, we're kind of figuring this out as we go along. That's the first thing. So, if you're confused and a little uncertain, that's the right place to be. I think we do know that the transformation involved in AI is going to hit more educated workers, more highly paid workers, more creative workers first, the kind of white collar professional work. But, it is important to recognize the difference between transformation and destruction. They're two different things. Your job is going to change, that doesn't necessarily mean your job is going to end.
Yes, some jobs may. I think we're all recognizing it. Look back to industrial automation and the industrial revolution. Some jobs absolutely went away. A lot of jobs transformed, and I think this is the interesting one. I'd love to get your thoughts on this. A lot of new jobs came into being that we couldn't imagine before some of those technological shifts, and I think we're going to see the same thing. There's going to be job descriptions three to four years from now, maybe less than that, that we don't even think are possible today. What do you think are some of the new things that are going to unfold for people?
So, I mean, part of what I think, the way we think about jobs, the way we analyze it, is jobs are actually bundles of tasks. You don't do one thing, you do many things. So, as CEO, you're doing podcasts, this is one part of what you do. You're doing leadership roles, you're writing everything, you have meetings, you keep up to date on the technology, you probably go on sales calls yourself, you talk to policy makers. Those are all tasks that you do. If AI takes or transforms some of those tasks, that will change the bundle of what you're doing at work. So, I have a feeling that it may be less job categories being created as much as people's jobs transforming about what parts of the job they specialize in.
No, that makes a lot of sense, because I think we're already seeing, like I think of it already in a lot of these kind of intelligence, white collar work, where I always say it's the first draft thing. Why would you create something from a clean sheet of paper today where you can give a few thoughts, get a first draft of a new blog post, a marketing paper, and let it do that initial first creation, and then bring your knowledge, intuition, whatever on top of that. That saves a lot of that up-front part of many, many tasks. That's certainly an obvious one to me.
So, I actually want to push back on two different ways on that, because I think there are two other ways. There's a risk of doing that, right? So, if the first draft goes to the AI, then you're not doing a lot of thinking, and you will end up following the AI's instructions. So, I actually recommend writing a very rough first draft yourself before handing anything to the AI, or you're just going to be saying the same things everyone else does, because the AI is going to give you fairly similar answers. And the second thing I would say is with the rise of agentic AI for real, and look, people have been talking about agents for the last few years. Agents became a real thing for most people in the last three to six weeks. And as a result of that, increasingly the way you want to work with AI is you just want to delegate work to it. So, it's not even saying, 'Oh, give me a first draft.' My book Co-Intelligence was all about how you work back and forth with AI. We're entering a world of AI management though, where you tell an AI to do something, it does something, and then it might save you seven hours or whatever of work.
And instead of sharing a task with it and having it enhance your production of the task, just give it the task, let it bring the work back to you, I guess would be one thought. But for creative work, you're saying that's kind of counterintuitive to what I've seen and read before. Better to use your initial thinking as an expert in some field to make sure your imprint starts it down a path, then let it enhance that.
That's right. Otherwise, you get fixated on a position. It's the same way with ideation. Actually, it turns out brainstorming is the worst technique for generating ideas as a group, because as soon as someone says something out loud, everyone fixates on that idea. And looks at the most senior person in the room. So, you actually always want to start a brainstorming session before AI with writing down your ideas privately first, so you didn't reveal all your thoughts in advance.
Interesting. Yeah, that's always true in a room when you have various hierarchy of decision makers. Everyone could say, 'Oh, we're all equal in this room,' but you're not. People can't separate themselves from the structure of the organization. And your point, you always say there's no bad ideas, let's get everything out there, but then almost people can't help themselves like, 'Oh, I don't like that thing.' And they start attacking it before you've even had a chance to, or they self-censor right at the beginning, right? So, the AI has that same kind of aspect, which is like, 'Hey, write me a blog post. These are good points. I'm going to focus on these points rather than the points you wrote down the path.' When you want to, it's really interesting. That's kind of counterintuitive for me. You've said, I think the quote was, it's going to make everybody good at work. Does that mean kind of take whatever we're doing and it's just going to enhance either our productivity or the quality of our work? What do you mean by that?
So, there's actually a lot to unpack. Some of our initial studies with my colleagues at Harvard, MIT, and University of Warwick, we went to Boston Consulting Group and we gave everyone access to GPT-4. Half of 8% of the workforce, half of them got access, half did not. We saw a 40% improvement in quality right away. And the lower performers had the biggest quality boost. Okay. And that one, so one way that AI makes people better is whatever you're worst at, the AI is probably better than you. So, it moves everybody up to the 80th or 90th percentile. So, that's one way it makes everybody good. It closes skill gaps. But in some ways that's the AI doing the work, not the humans. A second way to make you good is whatever you're best at, it probably works as an enhancer. So, if you're a good coder, you're probably a really good coder right now. So, it's both doing both things, skill compression of making everyone the same and also dispersing skills more.
So, I'd read something that was a little potentially counter to that and maybe it's an older piece of data now. I think I read something that early on, I think we thought, and this was particularly about software development and coding, that it would bring everybody up. Then I read something that said, 'No, actually it's going to take that really good coder and make them like ten times better than the average coder who might come up to be a good coder.' In other words, it was almost going to have more outsized positive impact on taking your great people to super great or whatever. And have you seen any of that kind of thinking at work?
So, there's really four things that are happening all at once. One is this idea of skill compression. Everybody, the floor gets raised for everybody. The second is that everybody gets elevated to some degree. And then there's the idea that maybe some of your best performers get very large boosts and there's some evidence for that as well. And then there's a fourth thing which we don't talk about as much, which is some people are just good at AI and they get a very large boost. So, people who were not necessarily the top performers before, maybe they're really good at agent decoding and suddenly they're producing 100 times more code than anyone else. So, all of those things are happening at the same time inside organizations.
That last one I had not heard much about and to your point, we remember that's long ago we were talking about digital natives. I'm sure pretty soon we're going to talk about AI natives, people who grew up using AI just in their normal course.
I actually worry about the digital natives analogy because when you were talking about digital native it was like there's a lot of stuff you have to know about making the internet work. So it's a check for you. So you know how the culture works, how language works. I think people are over-applying the idea of digital native to AI. Like one of the more depressing things I heard from a CHRO recently was that they said, 'Oh, the kids we just hired are great at AI.' I'm like, they're not great at AI, they're just typing everything you say into Claude and handing you the results back. Like that's not great at AI. Great at AI is not a useful skill.
What is the great at AI skill you think people are getting? If you don't have any experience, you don't have any knowledge, you're not great at, like, you're just handing an AI work. This is actually something where experienced workers have the biggest benefit. Well, that's a segue to one of the other places I wanted to go, which is one of the things I think a lot of people are wrestling with is so if AI is going to take over a lot of parts of our work, entire tasks on our behalf, where are we going to get the value from the humans? This is sort of your point, like if a young person right out of school has zero experience, but they throw things into AI, they don't have all this domain knowledge that would then theoretically allow them to use that AI more effectively. Someone who's been in the industry ten years doing whatever, software coding, investment banking. Talk about like what are we going to value in the humans as we get more and more dependent on AI.
All right, so this is sort of a bad news, uncertain news, good news situation. Okay, that's the order. So the bad news side is I think a lot of people want to have bright shining lines that AI can't do this. So I was talking to a really well-known social psychologist. He said when the AI first came out I thought it would be empathy and creativity would be human traits. It turns out in almost every survey we have, people would rather talk to GPT-4 than talk to a doctor because of its higher empathy levels. And so we're not finding those bright lines. So I think bright lines are a really dangerous place to be. On the good news side, AI is jagged. It's good at some stuff you wouldn't expect, bad at some stuff you wouldn't expect. And there is a need for humans to kind of fill lots of gaps about what AI can do. I don't know how much people will like what gaps there are. The gaps are going to be different in every field, but there's a reason you can't just say AI do my job. And there's a lot of pieces that are missing there. And then I think on the uncertain news is in the long term, what does this look like? When you have agents that you go assign work to, what does that mean for work? We just don't know. And anyone who tells you with certainty, like there's always some viral post one way or another, we just don't have answers.
Yes. Especially, and I think in the world we're kind of maybe move to next on some of the security aspects, agents interoperating with other agents without the new phrase humans in the loop. We talk a lot about AI with humans in the loop. I think that's sort of everybody's initial comfort zone, like there's still a person to check on it, make sure it's working. But pretty soon I think we all think there's going to be a level of agents interacting with other agents. And now that's completely unknown kind of a landscape for all of us, I would assume.
I mean, unknown for literally everybody on the planet. So the frontier of knowledge is not that far away from where you are right now. So anyone, if you by just saying agents interact with agents, you're already in the 0.001% of people who kind of understand what's happening in that space.
That's fair. Well, and let's go to some of this because again our company and a lot of the folks that listen in on this are folks that are dealing with enterprise companies. There's a ton of interesting things happening in the world of consumers and in the world of small business. But for the most part, SailPoint, a lot of the companies that we interact with, we're in that world of enterprises, from say mid-size, few thousand employees up to the giant corporations. And I know you've done a lot of work with those same kind of companies, the big brands we all know. In that realm, we're certainly hearing a lot about a lot of interest. I would love to hear, you said you've done some work with boards, I'd love to hear kind of what you're hearing in boardrooms about, well, we've got to step in and take advantage of this tech. And on the flip side, a lot of concerns about the safety, the security, the data protection. So just kind of what are you hearing in some of those business level conversations about concerns with security of data and access to data and those kind of things?
I mean, I talk to the heads of all the AI labs on a regular basis, not as much on the CEO side, but the people building the products. And they don't actually have that much of a security background. Like if you've used any of these enterprise tools, they don't have anything. They're just starting to add auditing and things that are basic stuff. And so they're trying to build the product that does everything. It's hard to have that happen. I would say on the security side, it's also kind of a mess right now. Part of the issue is that these tools are really effective. Right until very recently, it was all about co-intelligence, not to steal my book title, but working back and forth with the AI, which made it very easy to attribute problems or issues. So, you give someone the tool, they're responsible for the output of the tool. And the truth is the initial set of worries about what AI could do focused on stuff that wasn't really that much of a threat. Will it exfiltrate our data, will they train our data? That stuff has been solved problems for a while.
The more complicated issue of what does an agent have access to under what circumstances, that's a harder problem. Especially because the agents are pretty good at finding information and figuring out ways around that. And I think that people haven't grappled with that really yet. At the same time, when you can get large scale productivity gains, and when the survey work that my colleagues at work have been doing have been showing that now 75% of companies report positive ROI from their gen AI experiences. So, they're seeing gains from this. It makes it hard to say, okay, let's slow down and wait for a full solution. So, this is one of those building the planes in flight situations. But as the decision-making shifts from the CIO office, the tech office, the CISO office into the CTO, the CMO, the CEO's office, some of this is we just need to get this going. So, I think that part of this is when I see security people try and slow this down, it either results in a complete stop to everything, in which case the company starts falling behind on AI work, or they end up getting overruled. And I think that what we need to start seeing is a new kind of partnership, where the people worried about security and identity are doing this in a how do we enable as much as possible as quickly as possible, rather than how do we slow this down to be sane.
Totally get the impulse, right? There's all kinds of problems underway, but there's also a big risk in companies not adopting. You hit the word I wanted to hit next, so good segue. Well, the word our brains are tracking here, because I was going to bring up the word risk and say obviously what I think a lot of us are seeing is there's two kinds of risk. There's the risk of being too aggressive perhaps and not understanding some of the risks and then waltzing into something and all of a sudden all of your IP is all over the model and you've lost your IP. The other risk that we're not talking about as much is what about going too slow and getting lapped by your competitors who are more aggressively adopting it. How do you hear particularly senior exec teams or even boards discussing kind of those two risks?
They are less worried about the security risk than they were. I mean, part of that is look, you can go through the list of there's they have enterprise chat GPT at JP Morgan and Novo Nordisk. Like it's not like there's enough big players that have jumped that it's no longer quite the same kind of thing that it was a few months ago. And by the way, when you see big players jump, it's often because the CEO gets the religion. It doesn't take long of playing with one of these systems for people to go, oh my god, I have to figure something out here. Like the idea that you're not using agent coding tools is already ridiculous. Like of course you're going to, and we talked about Cursor and some of the IDE extensions, but the next step is Cloud Code or Codex where, and so how can you not if you get ten times or 100 times productivity gain? There's risks everywhere. I mean, you built this company, right? I'm an entrepreneurship professor, you're an entrepreneur. Part of what an entrepreneur does and why your entrepreneurial CEOs are so successful is they know when to take a risk and when not to, to balance it. You're not just a custodian, you're transforming. And I think the same thing happens here. Visionary CEOs have to take a risk. There's danger here, but there's danger everywhere right now.
Yeah, there's an old, you probably heard it somewhere along the way. We've repeated it on this podcast a few times just because it's such a simple but great metaphor, which is why do F1 race cars have great brakes? Answer, so they can go even faster. If you trust your brakes, you can come into a turn at 150 miles an hour, brake to 80, navigate that turn, and get right back up to 150. You need a great engine, that's the ability to take advantage of this technology, but there is this sense of I want to trust that I have a brake when I need it. And that is risk management. Like the most successful entrepreneurs weren't necessarily risk-blind. They took reasonable risks, sometimes pretty edgy risk. But then they also understood the risks that were not okay to take to put the business. And I think that we're in that unknown area right now.
