ADI IGNATIUS: I’m Adi Ignatius.
ALISON BEARD: I’m Alison Beard, and this is the HBR IdeaCast.
ADI IGNATIUS: So Alison, we are at week three of our special series on AI. And after diving deep on talent and communication, we are looking today at innovation.
ALISON BEARD: I’m excited about this one because there’s no question that generative AI is itself the most important technological innovation we’ve seen in decades, maybe our lifetimes, and that it’s supercharging innovation in every other realm of business from cancer treatments to space exploration. But I guess there’s still a question if all of the innovation is that impressive or differentiating when everyone has access to the same tools.
ADI IGNATIUS: Yeah, so that’s a great question. Look, I’m still trying to understand what an AI-powered company really is, how the technology can create not just efficiencies, but can transform businesses and allow them to innovate at a scale that would’ve once been unimaginable. So to help explain all of this, my guest today is Jeetu Patel, who’s leading AI strategy at Cisco, a company that’s helping to build the infrastructure for the AI era. He has a lot to say about how AI can help drive innovation and how companies need to think about processes, about security, and about preserving human qualities like judgment. Here’s our conversation.
AI is dramatically reducing the time and the cost required to build, test, iterate on ideas. In your mind, does that fundamentally change what innovation is from a scarce activity to something that we should think about as almost continuous?
JEETU PATEL: When there’s an element of abundance where you can actually start thinking and prosecuting ideas at a very fast pace, the rationing that you have to do is making sure that you don’t create slop. In a world where you can write unlimited amounts of code, choosing what not to build is actually a far more important decision than deciding what to build.
And from that perspective, I do feel like the need for exercising human judgment and taste and intuition and instinct is actually more important now than ever before. And so I always find it interesting when people say, “Well, all humans are going to lose their jobs or we’re not going to have anything to do.” I just don’t even understand that because there’s going to be so many areas where human instinct is going to be more required than less as we move forward.
ADI IGNATIUS: Well, let’s talk about processes a little bit because our companies have spent decades creating and perfecting this whole process for innovation, roadmaps, planning, approval processes. Which of these processes should become obsolete when you can go from an idea to a prototype in hours, in minutes, instead of months?
JEETU PATEL: The way that I think about processes in general is you have to be not pedantic about the process and you have to be far more intellectually honest about the level of rigor that is done to go out and drive to a certain outcome and that you’re sweating the details.
Let me give you a very simple example. Have you seen someone create a PowerPoint in Claude or OpenAI and then start to present it to you? I find those to be less effective than when someone has actually taken the time to build the PowerPoint.
Something must happen in the human brain where people are so eager on hitting send without actually going through and saying, “Is there thoughtful analysis done? Is the point being made in the right way?”
What I found is the quality of conversations were starting to go down because even the people who had authored the document weren’t the ones who were intimately familiar with every single concept in the document because they were just kind of regurgitating what someone else had created. I think that’s a bad way to do things.
The goal of AI is to not degrade our thinking, but to enhance our thinking and allow us to think at a broader and higher aperture than what we were able to do before. And I don’t think that that’s something that we have adjusted to.
The most important thing we need to do is can we just improve? If there were three things that we were doing in a day, sure, one thing is can I do 10 things in a day, but that doesn’t really change the world as much as saying those three things that I’m doing, can I do them 100 X better than what I was doing before? And that would actually create a much better outcome.
We’re moving from chat bots to these agents that can intelligently answer our questions and conduct tasks and jobs for us almost fully autonomously. But it was very hard to go out and use agents. What is now starting to happen is these personal agents like Grokbot, Muse, and Instinct are going to democratize that quite dramatically. And if you leave aside the fear on safety and security, which we should talk about, and if you said, “I can enrich those agents with the ultimate amount of context about me, can that agent really in a meaningful way lower my burden and help me out and actually make me feel like I can produce better quality thinking for myself because of the agent aiding me?”
I think the answer is categorically yes. And that is an area that I feel is going to be the next very exciting phase in AI is these personal agents that we’re using that’ll then bleed into the business agents as well.
ADI IGNATIUS: At this point, is AI generally for companies, for teams, still essentially making things more efficient or is it changing the kinds of problems that we can choose to tackle? Is AI changing how we even think about problems we can solve?
JEETU PATEL: I think that’s the more interesting part rather than the efficiency part. The efficiency will come. But the thing that is truly the force multiplier is original insights that get generated by AI that you would’ve never been able to generate without it, that then allow you to expand your thinking and your perspectives, and then that allows you to have the ambition and the credibility to be able to solve problems that you could never even dream of solving before.
Especially in the United States, we’ve gotten very negative on the narrative on AI, and I actually don’t like that because I think this is a tremendous technology that can really help in solving problems across a multitude of dimensions that are going to be really great. And we should also, of course, be focused on mitigating the risk.
But this constant level of negativity around AI, I don’t think is accurately reflective of the potential of the technology. We are using AI now more than any time before in our history. 32,000 engineers are coding with AI. We will actually, by the end of next year, have 70% of our products that are 100% written with AI. Interestingly enough, we will have more engineers than less at the end of next year at Cisco.
You’ll need more people to think differently about things that we will be able to do. And by the way, every piece of code that an AI agent writes should have human provenance to it, that it’s actually tied back to a person that owns the responsibility of that. Don’t just delegate it to AI and say, let’s see what happens.
ADI IGNATIUS: Well, but that’s an interesting question, because in my mind, if AI is going to achieve these sort of remarkable step change innovations, at a certain point we have to trust the output. What’s your suggestion to when we can have a comfort level with AI making consequential decisions and we don’t have to check everything?
JEETU PATEL: I think it depends on the domain. There’s a technological reality and then there’s a psychological reality. The hallucination factor with AI has gone down quite materially since 2022 when ChatGPT first came out. The public sentiment on hallucination has not improved as much as the technology has improved.
ADI IGNATIUS: First impressions.
JEETU PATEL: Right? Part of this is you will actually need to make sure that you see the degree of improvement that happens technologically. We will need to make sure that that actually is effectively reflected in the human psychology of it. And when both of those things happen, that is when society will be ready to say, “I’m going to be ready to give it consequential decisions.” Today, we are not 100% in either of those. There’s a lot of risks. I think there’s a lot of talk around these agents going rogue or doing things in ways that might actually exercise poor judgment. The way I’ve always talked about this, agents are like teenagers, they’re supremely intelligent, they have no fear of consequence, and they exercise poor decisions far more frequently than you and I would like.
ADI IGNATIUS: Especially, we learned, when you put them in a room together.
JEETU PATEL: That’s right, that’s right. So I do feel like we need to make sure that we actually solve those problems and we have the right level of guardrails and we’ve got the right level of behavior monitoring, and we make sure that we can determine when something is wrong because there’s lack of alignment with what we want to do, which alignment basically is a fancy way of saying, me, the human, what I want out of AI, is it going to follow my instructions to do what I want it to do?
We are in the early days of getting those problems solved and we need to make sure that we work on it fast, but you have to know that these technologies move exponentially and they move much faster than we might think.
And second is the psychological issue that we have to solve, which is are people feeling a level of comfort with this as the technology gets better and we have to make sure that we actually get people educated as the technology is getting better.
ADI IGNATIUS: So on the security issue, the idea of rogue agents doing scary things really went from worst case scenario, sci-fi to, oh, it’s happening and most, if not all, of the frontier models have experienced this. So we could say, look, we hope the frontier models clean up their game. We hope governments come up with sensible regulation. But for companies, for executives, what should they do internally to mitigate the security risks that we’ve been talking about?
JEETU PATEL: I think you have to have a very, very thoughtful and diligent security practice within every company because these can be business ending events and you have to make sure that you are assessing the risks and you’ve actually got technologies that can intercept at what they call runtime, which is at the time that these things are happening rather than post-facto so that you can stop an agent.
It’s kind of like going back to the teenager analogy, you would not let a 13-year-old or a 16-year-old who’s just learned to drive be given a very fast sports car to say, “Go do whatever you want. If you want to go out and drink and then come back home driving, everything is okay.” You would never do that. You would actually have guardrails in them. You would make sure that your Life360 can track how fast your kid is going, and if you start to see drift from the expected behavior that is not aligned with you, you would actually take corrective action right away because it’s your child.
I don’t think AI is that much different from that. You have to manage these agents, but you have to do those not by humans overseeing it, you’re going to have to have that happen at machine scale because it’s not going to happen at human scale because the volume’s going to be so large. You’re going to have trillions of agents in the world. These trillions of agents are going to need some kind of monitoring ability that is also going to happen with agents and you’ll have a check and balance between agents. You will have models that’ll be checked for safety and security by other models, and you will see partnerships that are happening among competitors because they will all kind of keep each other honest.
Elon had suggested this and it was a really good idea, is have each one of these models test the other model for safety and security and then make sure that you publish the results and have the harnesses be open sourced. And then that’ll actually start to see it creating benefits in a way that is good for humanity as a whole.
ADI IGNATIUS: All right, so let’s put on our optimistic hats and let’s imagine that we solve some of these problems. So I’m still trying to imagine a company that is genuinely AI native, not just using copilots and achieving efficiency, but one in which, I don’t know, hundreds of thousands of agents are continually experimenting, building, analyzing, acting. What does the management system of that company look like and what is that company suddenly able to do?
JEETU PATEL: Well, I’ll give you our example. We have 90,000 agents that we’ve rolled out to every employee. There’s an agent for every employee, and that agent, it has access to my email, it has access to my messages. We’ve built our own harness for security around that agent. We monitor how the agents are behaving, and we basically ask the agent to say…
On a weekly basis, I have the agent just give me, rather than me going out and actively looking at systems for how our numbers are doing, I tell the agent to say, “Give me a summary of how we did in our financials last quarter on both revenue and top line as well as on the budget.” I asked the agent to say, “During my course of the day, if I’ve committed things to other people, give me a list of the things I’ve committed to at the end of the day by listening to my meetings, by making sure that you look at my messages so that I know what I have to make sure that I come through for delivering on those commitments.” And so it actually becomes this aide and assistant for me.
I brainstorm with it. I learn with it. I actually do deep research with it. It’s fantastic. You have to think about these agents like digital coworkers that have been augmented to your team, and you just have to make sure that all of us are going to be managers of agents, and we have to learn how to use that agent force more effectively than the way that we have used it in the past. Because you’re going to have a human force and you’re going to have an agent force, and the combination of those two will allow you to get from point A to point B.
ADI IGNATIUS: So if we all have access to those tools, if experimentation becomes easy, nearly free, what is competitive advantage? I mean, you said judgment and that’s something, but it seems you need an organizational ability to do something to test and learn and scale faster than anyone – how do you think about competitive advantage in a world where everyone has these amazing tools?
JEETU PATEL: What is competitive advantage? Let’s say that there’s two companies with a thousand people. They’re making the exact same product for the exact same market. Why is it that one company does better than the other? What do they do? The decisions they make on how to resource and focus the thousand people really matters. And so two companies can have entirely different outcomes being in the same business.
Agents are no different. Just because we have access to these resources doesn’t mean that we are going to utilize them in the most effective way possible. That still requires human instinct and judgment and intuition. And so competitive advantage is, how effectively can you deploy the resources available to you to go produce an outcome that is going to be a superior outcome in the market that is highly differentiated? That is what defines competitive advantage.
That doesn’t change with the fact that you have availability of agents, that just means that now you have another resource to tap into so that you can go out and deliver an outcome that is meaningfully differentiated, that is value creating, that can allow people to do things that they weren’t thinking of doing before, that can help solve problems that are important for people. That is what creates competitive advantage, not the ability to have the resource.
ADI IGNATIUS: So what happens then to the traditional innovation team? If marketers, engineers, designers, planners, whatever, can all use agents to prototype and build, do the boundaries between those roles start to disappear? Do we need to rethink the roles that we even have within a corporation?
JEETU PATEL: Firstly, all of them get supercharged is the way to think about it. And two is the roles, almost every job will get reconfigured. Some jobs will be lost. An entirely new set of jobs will get created. By the way, this is no different from any other technological revolution that’s been had in the past, however, this one seems a little bit more uncomfortable because of the speed and pace at which things are moving.
I think we live in the bubble in Silicon Valley and it’s so profoundly insular in some ways because everyone believes in Silicon Valley that whatever you are doing on a daily basis is the norm all around the world and it so isn’t. That’s why I find it superbly valuable to travel and go to different parts of the world and experience what people are experiencing there and saying, “Okay, so what is the gap in the delta and what do we need to do to make sure that this technology gets diffused evenly?”
But in my mind, the overarching thing that we have to keep in mind is we now have intelligence that is made available to us with no level of scarcity, and it is up to us to make sure that we direct the intelligence in the right way so that we can get the right levels of outcomes from it. But that directing of the intelligence is actually a non-trivial task that I think is going to take a long time to get automated.
ADI IGNATIUS: As this is all rolling out, as the potential is there to grab, what’s the biggest leadership mistake that you see companies making in terms of AI’s capability, in terms of the risk of AI?
JEETU PATEL: I think one of the big mistakes that people make is thinking that just because you have some tools that all of a sudden you will just be able to 10X your throughput. It’s not about 10Xing your throughput, you will have to make sure that you actually create better outcomes, but it is going to require… Every single time there’s a step function improvement in one dimension, there’s a human bottleneck that has emerged in another dimension. And that’s what I’m seeing so far.
Coding is largely a solved problem at this point and has gotten automated quite a bit. Code review becomes a bottleneck. You will actually eventually solve code review as a bottleneck and then deciding what to build and what not to build is going to become a bottleneck.
There’s this kind of magical thing that happens when there’s a light bulb that goes off because of serendipity or because you actually talk to people and there’s a spark of insight that comes out. Now what’s going to happen is those sparks of insight will be more frequent because you will be able to do that not just when you’re physically with someone, but you’ll also be able to do it in a very, very kind of intentional way with AI.
But it requires that you actually incorporate that into your day-to-day workflow, and if you don’t incorporate it, then you don’t get any benefit from it. Then whatever it’s giving you… This is why I say, asking AI to do a bunch of work for you and then not reviewing the work and not internalizing the work doesn’t really help any. The whole point is for it to do something that you can then consume and do something with, and when you can make that happen, that’s when magic starts to occur.
ADI IGNATIUS: So I think at Cisco under the leadership of you and others is very AI forward. Talk about what has actually changed inside Cisco, whether it’s budgeting, product development, incentives, decision rights aimed at increasing the pace of innovation at this moment.
JEETU PATEL: And I think it’s important to also understand what’s not changed. And so what’s changed? The way in which we think about building products has completely changed. The way in which we position and market products has completely changed. The way in which we think about adding value to customers in the customer meeting and selling products has completely changed. The way in which we manage and operate the business has completely started to change. Those things have changed quite dramatically and quite materially.
The way in which we learn as individuals and collectively as teams has changed and the velocity has increased quite a bit. The way in which we do research has changed. The way in which certain workflows get carried out and get automated has completely changed.
What has not changed is the desire for people to want to go out and contribute and add value, the desire for people to succeed, the need for status, the need for recognition, those things haven’t changed in people. People are still the same. They still need those things.
And I think that’s a pretty important dynamic to keep in mind as we think through this. The core foundation of what humans want, those basics, I think are going to be remarkably similar 20 years from now. People still want to feel like they’re loved. They’re going to feel like they’re being valued. They’re going to want to feel like they’re engaged in something purposeful and larger than themselves in life. They want to feel like they have a comfortable living and a comfortable life. They want to feel like they’ve actually progressed and done well in society, they’ve contributed. All of those things will remain the same.
ADI IGNATIUS: Is there something Cisco can now build in say a week that would’ve taken a year, two, three years ago?
JEETU PATEL: I’ll give you a very concrete example. One of our biggest problems that we had was over time—$63 billion in revenue, we made 250 acquisitions, the way in which you manage all this infrastructure was multiple different products that all looked and felt different. We felt like a holding company. We decided to become an integrated platform company. So we said we’re going to have one management plane for all of our products, which it was an absurd ambition to start with, completely absurd ambition to start with.
And so when I brought this up to the team, they said, “Well, Jeetu, this would take, the way that you’re describing it probably, I don’t know, six, seven years to go out and build in the right way and you’d need all of these resources.” We built it with a small but mighty team and made it everyone’s priority to help that team by adding to it within a matter of nine months. Something that was going to take seven, eight, nine years, we were able to build within nine months.
Because we were able to move at a pace that was just not even imaginable that we would be able to move at. That’s just an example – the clock speed of a business changes dramatically. And by the way, it’s very hard to get adjusted to that clock speed of the business because it’s cognitively disorienting because what are you talking about? That can’t be, it’s not going to happen that fast. There must be some kind of catch to this. It’s like, no, it’s just you’ve got machines now writing code and that’s largely gotten automated, and as a result, you can instruct it to write any code and it’ll probably write it for you pretty fast.
ADI IGNATIUS: So you talked about bottlenecks to innovation. What has proven hardest to change, something that needs to evolve, but it’s hard whether it is for these human reasons or maybe technological reasons?
JEETU PATEL: Hardest to change is culture. I feel like the cultural aspects of this, of getting people to feel safe about learning and experimentation, getting people to make sure that they can be creative in the way in which they go about doing these pieces, I think that takes a while in organizations.
So first, I’ll tell you this, at Cisco, the first six months I really struggled with getting everyone to start using it. And then a light bulb went off with a few people and they started, and the technology matured during that time quite materially as well. The one thing that’s a failure state that happens in AI is let’s say you use AI and you don’t find a great experience.
Let’s say that you got started and you use it today, and you don’t really find it to be earth-shattering. The typical behavior that a human would have is say, okay, I tried it out. It didn’t work. I’m going to put it away for a while. I’m going to put it away for a while. It can tend to be six months, can tend to be 12 months. And you’re like, “Yeah, that didn’t really work for me.”
The reality is you have to assume that the things that didn’t work for you will actually start working for you in the next two to three months, and as a result, you have to assume that if that worked, would that be a really, really good thing for you? And you have to start planning while it’s not working that it is eventually going to work. And I feel like that is the hardest mental model shift that people have because they’re like, “Yeah, I tried it out. It didn’t work.”
But this thing is moving at an exponential curve. The exponential curve is almost a vertical line. That’s how fast this thing is moving. And so from the time that you think about something to the time that that thing actually got incorporated into the product is a matter of minutes. And so try again next week and you might actually find yourself surprised.
Right now, my guess is sub 2% of humans are- using agents in any kind of substantive capacity in the world. Imagine what happens when 10%, 20%, 30%, 40%, 50% of humanity starts using it. You can start to see a very, very different level of productivity ratio. Standards of living will go up.
And we will become smarter beings as a result of AI being around us and we will therefore be able to solve problems, cure diseases, reduce poverty, diffuse education in a much more broad way in ways that we would never have imagined we would be able to do.
ADI IGNATIUS: That’s great. Jeetu, thank you for being on IdeaCast.
JEETU PATEL: Thank you for having me.
ADI IGNATIUS: That was Jeetu Patel, Chief Product Officer at Cisco, a company that is helping to build the infrastructure for AI. We’ll be back with a regular episode on Tuesday and then check back in your feed on Thursday as we conclude our AI series with a conversation with Wharton’s Ethan Mollick on AI and strategy.
If you found this episode helpful, please share it with a colleague and be sure to subscribe and rate IdeaCast in Apple Podcasts, Spotify, or wherever you listen. If you want to help leaders move the world forward, please consider subscribing to Harvard Business Review. You’ll get access to the HBR mobile app, the weekly exclusive insider newsletter, and unlimited access to HBR online. Just head to hbr.org/subscribe.
Thanks to our team, senior producer Mary Dooe, and senior production editor Kristin Murphy Romano, and thanks to you for listening to the HBR IdeaCast. We’ll be back with a new regular episode on Tuesday. I’m Adi Ignatius.