How Mihir Shukla Is Reimagining Work With 300 Million AI Agents (and Counting)

0:00I'm Mihir Shukla, CEO and chairman of Automation Anywhere. Automation Anywhere is a software company. We help customers create AI agents and automation on our platform. We have over 5 ,000 customers in 90 countries, some of the largest companies on the planet, too many mid-sized customers use us. We have 300 million AI agents running on our platform, and this has been doubling at a rapid pace. So at one point, you are looking at a billion AI agents on our platform. Anybody thinks that future is yet to come, it is here and it is accelerating fast. But what we are about is trying to reimagine how work happens.

0:43How much of work can happen autonomously so that we can focus on work that matters most? How much work AI can assist you so that you can be more creative? And what part of work really needs to be manual in the year 2025? We're really, really thrilled. Maybe just by way of introduction, Mahir, can you talk for a second about your, I know you're undertaking a pretty broad transformation effort at Automation Anywhere. Tell us what are the kind of key elements of your vision that you are seeking as you're pushing forward into this transformation? So we have the vision that we thought with AI, it is possible to run a large part of enterprises in an autonomous way.

1:31And so the question is, what percentage of your enterprise will run autonomously? What percentage will run assisted by AI? And what percentage will be manual? And we've given ourselves a license to rethink what that looks like. So we are doing this for 5 ,000 plus of our customers and ourselves to reimagine how work could happen very differently. It turns out in places like customer service, you could have 40 % of customer ticket handled autonomously, no human in the loop, within a matter of seconds. in places like tax and certain finance operation, over 90 % can be autonomous. So it varies by functions.

2:21Yeah, I think I saw the CEO of Goldman say something like 95 % of a prospectus can be written with AI, right? That's correct. I have so many questions. Jeremy, what do you want to start? Tamimah here, I know we talked about this as well earlier. Start with your organization itself, because it's one thing to go tell other people to do something. It's another thing to say, here's what we've done. Can you talk about the organization itself? Because I would say you are a leading indicator of what's possible for others. How did you start by assessing your own organization? What were some of the early areas of focus, early wins that led you to believe?

2:59There's actually something here for our customers as well. Happy to. So we try to reimagine work across every single aspect of our work. So let's take an example of our customer service. We provide enterprise software to some of the largest and mid-sized companies on the planet. And so when we deployed our software on customer service, we found out that 40 % of customer requests didn't need a human in the load. There are certain kinds of workload where, let's say, if you're trying to figure out what is the status of my ticket, You don't always want to talk to somebody if you can get a status right away.

3:40So about 40 % of tickets could be completely handled. And it lifts the level of quality of service we could provide because now people could focus a lot more on the other tickets that needed more time. Take example of tax operations. Tax operation for the highest level is applying set of rules to a set of numbers. there is always an aspect of strategic tax view that you can apply, but that's about 10%. That 90 % could be fully automated. So when you use those two examples, customer service and tax, for example, I'm thinking in terms of principles others can leverage. And I don't think it's necessarily, look at your customer service, look at your tax, right?

4:23It's more, what were the principles that led you to those areas? Were you thinking, where is our operation heavily rule-based, for example? I'm making this up, right, just based on where I'm listening to you. But what were the selection criteria that led you to customer service or tax for your own business first? So I think the way to think is that there are two different criteria as we applied. Of course, the rule-based is very easy to automate. Even though rules like tax rules can be 185 ,000 pages long, with the power of LLMs, now it is possible to understand all of those rules and apply it to set of numbers.

5:02But the other way to think is that where there are a large number of knowledge workers and wherever there are a large number of knowledge workers, what we have done in our industrial way of scaling is we have defined standard operating procedures so that we can produce consistent quality. That means like we have, let's say we have 500 customer service agents. We have a way to operate it in a more standard way, which means there is a significant room for automations there. So if you take a look at our customers, let's say if you are a healthcare customer, claims management, millions and billions of claims that come in their way could be significantly automated.

5:51If you are a bank, mortgage applications, KYC, can be significantly automated. A huge part of procurement processes, a significant part of HR operations. Maybe I can ask you on the more the social engineering side of things. I think a lot of CEOs and boards and senior members are sitting and having, obviously, these conversations. And then you go down to, let's say, your finance team, and you're saying, I've heard that 90 % of your operation can be automated with Gen AI. And then your CFO might say, yeah, but like ours is very complicated. And like, it seems to me that increasingly the technology is increasingly there and available.

6:41And what seems to be kind of lacking is the social engineering tools, all the processes, to kind of help us adapt organizations that was designed for a time of internet to now time of AI. How do you think about that? I think first, it seems to talk about why should we do this. Just because AI is available doesn't mean we have to use it. It's a tool, a technology. So maybe if you start with a why, and if once it becomes clear to you that this needs to be done, then often human beings find a way, right? So there are two reasons. I always start with the why, because what and how people can often figure out if they're clear about why.

7:25I think there are two reasons why we have to use AI. One is that if you take a look at stats, for many, many years, every decade, our productivity was improving. So it used to take eight people to produce a million dollar, then it took six people. And then in 2008, it took 5.1 % to produce a million dollar. Since 2008, that hasn't changed. In 2024-25, it still takes 5.1 % to produce a million dollar, which means in the last 16 years, we have spent collectively on the planet over$40 trillion on IT spend, and we haven't improved their productivity at all. That's a problem. If you keep doing it, their business will not improve.

8:18Especially with aging workforces, right? Especially with population decline, right? If we aren't finding a way to use that number, yeah, I mean, I didn't mean to steal your thunder there, but my mind immediately goes to, but quickly it becomes an existential necessity, right? Exactly. That is the second piece with a 3 % declining workforce. So for the last 2000 years, the economy has been a twin plane engine. One productivity increase, another is the more people working in the workforce. On one hand, we haven't improved the productivity as much. On second hand, it's a declining workforce. First time in human history, we are living in a society where you have globally 3 % declining workforce.

8:58And it is going to remain that way for a foreseeable future. So how do you operate an economy? How do you operate in a world where there are less people working every year? This is entirely new for everybody. And what happens when we can't figure out? So the fundamentals of this is that we have to double the productivity that we have produced in last many years. Now think about it, the internet, computers, every software ever returned. It took so many things, so many innovations, the iPhone, everything combined, produce the productivity where we are. Does anybody have an idea how to double it?

9:38Wait, but you said it hasn't changed since 2008, which if my memory serves, I was in business school at that time and I didn't have an iPhone. Yeah. Which is to say the introduction of the iPhone and social media and all of Call It Web 2, this is a question to you. It did not move the productivity needle. Is that correct? It did not move the productivity needle. For businesses, you know, we might have our personal life got better, but in businesses, it still takes 5.1 % to produce a million dollars. That hasn't changed. Wow. Yeah. That's a staggering realization, actually. That's right. Because I would say there's this kind of, I get choked up when I think about it.

10:16There's this kind of almost delusion of productivity, right? The fact that we've got our iPhone, we're going to bed with these devices by our beds. We wake up, we look at them first thing in the morning. I'm sure the implicit rationale, even though none of us would say it this way, is we're being far more productive. We're producing far more value. But what you just said, since 2008, the amount of kind of dollars per capita productivity have not improved despite all of the advances to our quote unquote productivity suite. Is that right? That's correct. And it is because the software that we use is, you know, by the time you implement some of the enterprise software, it takes a few years to implement them fully and the world has changed.

11:01So by the time you start using it, all the benefits that you had thought would come, you don't realize all of them. And then very soon, world changes even more. And now you're behind in how your software is configured versus how world is configured. You know what I think of? I mean, there's kind of a direct impact and there's also it's analogous. But a lot of the research, you look at, you know, what Cal Newport, for example, great researcher, wrote a book about slow productivity. But one of the things that he highlights is the cost of distractions. And for every, you know, ping or notification, we lose something like 15 minutes of focus, right?

11:36For every time your phone vibrates and say, I'm right. Last night I was, you know, up late writing a blog post because I was distracted during the day. I wasn't able to do it. Basically for every, you know, ping that comes in while I'm doing that focus work, I lose 15 minutes of productivity. And certainly there's kind of a direct implication there to productivity. But I also wonder to your point about organizational renewal and integrations, if you could think about kind of software updates as effectively distractions or effectively notifications to the organization. And every time the organization has to update or adopt a new technology, it's the effective productivity impact of a distraction on an individual worker.

12:16What do you think about that? Yeah. I think we still work in a World War II task-oriented era, and we think replying email or replying messages is work. We are studying how work gets done, and it takes 32 emails, 82 messages, six meetings, and three years later, you're going to get your work done. We are the viewers of our generation because each of those is a thread in how we are viewing work, right? And now when you sit in front of an AI system and if you get your prompt right, you get the whole cloth out, right? Instead of step by step. So if that analogy makes sense, you are looking at a very different way work gets done.

13:03Look, where I was going with though is, just to finish the thought on why, is if we fail to solve these problems that we talked about, about declining population and lack of productivity, what happens is there is less to go around. And when there is less to go around, the society cannot support its infrastructure. It cannot support bridges, cannot support health care, cannot support social security. The democracies weaken. There are more riots on the street. There are more wars on the planet. Does all of this sound familiar? There is more at stake here than just use AI. Look, this is a technology.

13:48If there was a new energy source or a new metal or something new that would help solve the problem of this, we should use it. It's about solving a problem. I think AI is coming at an opportune moment to solve a larger societal challenges. It's a tool like we have harnessed horses and winds and nuclear energy. It's power to be harnessed. But just because you've been through it, I'm kind of keen to try to kind of like unpack a little bit more the practicality of it. Because obviously it makes a lot of sense kind of starting with the why. And the why can range from everything to like we need to kind of improve our EBIT to you won't be able to hire more people in your team and you're getting more and more work on your plate to, you know, we need to kind of create productivity for the world because otherwise all these negative consequences have it.

14:39But still, I find going then into your finance team, there's still kind of like, I think, this kind of invisible wall between the technology that is available today and then even what seems to be kind of like a real intent for people to do it. Have you found a way of kind of breaking down this invisible wall maybe in your organization or for some of the customers that you work with? That's a great question, Henrik Because often the first thing you hear from people is that No, but my work is very unique Or what we do is so special What we found is that if you get people started in this direction If you have 10 people in a group You will find that two people will always be looking for a different, better way to work In general, that has been true across customers and they will get you started by doing something.

15:36Did you say two out of 10? Is that the ratio, two out of every 10 people? Two out of 10 people, could be three out of 10. And then the other four or five people will see that, see the possibility, then they get excited about it. They won't be the first one to start, but they see it and now they can imagine the possibility having seen it. And then it catches like a wildfire. you'll always have one or two people in the room who only use the new smartphone because the old phones are not available. So there are always people who will resist change. But in general, you could get 80, 85 % of the organization moving if you get them started.

16:18I think the key is to, instead of change happening to them, the key is the teams leading the change. That's the key to bringing change. Yeah. And that's actually dovetails perfectly with where I was curious to go, which is what are the organizational structural implications? What do businesses need to do to reorganize? And you kind of hinted at it that teams may need to lead the change, but what does that look like practically? Yeah. So the way many of our customers do is they start by challenging the team and say, look, I want 10 % of work to be fully autonomous in a department or 20 % fully autonomous.

17:01And often people start with some funding available because otherwise you get stuck in funding and you stop seeing the larger possibility. it. And as I said, there are always be team members who are excited about new ways of working. Once you get to a certain outcome in six months or a year, then the CFO goes back to the team and say, look, since you got to the 20 % autonomous, can I get 10 % return back in my financial? And you get to keep the 10%, but I need 10 % on my EBITDA. And then 10 % is reinvested back in the business. Slowly, you begin to negotiate with teams. Once they have seen the possibility, they would resist less.

17:45Sometimes when you start with it, you could face a resistance because it looks like a way to cut people. So better to engage people on art of the possible before you begin to restructure teams and the possibilities. ahead. So what I'm hearing you say is there needs to be an ambitious goal, something like 10 % autonomous, you know, uh, in any given function, then you've got to provide some funding, I suppose, for exploration, for training, for experimentation, et cetera. And then there's, you know, an outcome that then becomes a negotiation with the CFO. Is it sufficient for a typical team to have the goal and have the funding?

18:29Are they off to the races or do you find, is there, and maybe this loaded question. I find that giving folks permission is one thing, equipping them and enabling them with not only psychological safety, but also actual pragmatic skills. If you told me, Jeremy, go make a song, I mean, other than with AI, which now I can, the problem is I don't know any instruments. So you can say, and you can say, hey, I'm going to give you$10 million. Unless I invest in my ability to do that, and you can give me a year, I'm not coming back with a song, So how do you think about other elements besides call it a goal, funding, or is your observation, Ben, that that's sufficient to tip the people, as you said, the two out of 10 who are eager to make a change?

19:13Is that enough for them? I think you pointed out a key element, which is to train them on some of the tools and capabilities on what is possible. So at Automation Anywhere, we have trained over a million people on how to use these capabilities all over the world in about 90 countries. And the gist of one week, there is a basic training available for one week. And that one week training opens up your eyes to what is possible. Even beings are amazing. They're amazing everywhere. Once you show them that, you know, your wills start churning and everybody comes up with ways to do things in their respective teams and departments.

19:59We have today 300 million processes running on our platform. Can you imagine? This is all innovative thinking by our customers and partners, right? Who knew 300 million things could be automated? People are very resourceful. To your point, Jeremy, once you train them and give them some funding and tell them why this is so important. How much do you think if productivity hasn't changed since 2008, what's your gut feel on how long it'll take AI to make a dent into that stat and how impactful it'll be? So I think, as I said, today it takes 5.1 % to produce a million dollar. In another 10 years, you could get to three, which is where we should be in a normal course of action.

20:55I think our current state is stagnated. But if we continue, you know, last 70 years of our journey towards getting work done better, that's what you're looking at. Does the technology have the capability? Absolutely. Change is hard for people. So we'll have to rally around us, reorganize ourselves around it and, you know, use this amazing capabilities to get there. Well, you talk, yeah, I mean, it sounds like you've been studying, as you said earlier, I was just looking back through my notes, you said, we've been studying how work is done. Yeah. And you kind of characterized it a second ago. I'd love for you to tell us first, what have you found as you studied how work is done?

21:35And then second, how are you rethinking how work should be done? Yeah. So we observed that a couple of points. One is that when we think of this enterprise applications, whether it is your ERP applications or CRM applications or HR applications or many others, we found that a user primarily spends 15 % of their time in that application. And they use about 8 to 18 different applications to get worked in. So the nature of work is very different than what we might believe. So you are spending far more time outside of your core application, across application to kind of go from Outlook to Excel to ERP to something else to something else to orchestrate a work.

22:27That's the nature of the work that has become. And as a result, it takes a lot longer to get work done. The second piece I mentioned is that we are kind of weaving the work. We confuse work with the task. This is the World War. Since World War II mindset, task is a work. So sending an email and sending messages, going to meetings. But one of the things that we do this experiment with people and say, what if instead of doing all of this, you have to finish this work and produce end outcome in three steps. And when people were faced with it, eventually they figured out how to not do all of those things and find a way to get it done in three steps, right?

23:14Like a shortcut. You know, you don't have to send 32 emails to get to the outcomes always. Sometimes there is a place for it. So we confuse tasks with the work, right? And the reason why this is very important is when AI will get to it in one step and a couple of prompt changes and you get the outcome, we better begin to think about it differently. The way of thinking about our work has to fundamentally change. I was just going to mention that I saw this happening in a few workplaces, including ours, where people who got better at prompt engineering with AI system begin to ask a different question in the meeting because they begin to expect the same out of human being and say, I'm not asking you, did you send the email?

24:03Because that's not what I ask in AI system. I'm not asking a task. I'm saying, when can I get this outcome that I get? Right. Can I get that? Can I get that? um to be very interesting on that point i'm increasingly in a quite a crazy way looking at the organizations i'm involved in in the same way where everything is an input and output everything is a prompt and a response and the prompt might be to a human or it might be to an ai it doesn't really matter but if you start to actually look at your organization structure as just api calls it kind of actually changed a little bit how you think about it.

24:44I did have another question though. One is, you know, you through the organization are such a good student of how do you take repetitive kind of workflows and then basically replace it with a bot. When you personally are looking at your own work, I imagine that in spite of the processes you have there, you started to do that. Could you mention a few examples of how you kind of look at your own personal repetitive workflows and then where do you use Gen.ai to reduce that workload? It has been eye-opening, Henrik, to that question. So take an example. My team has made a specific LLM for the CEO and for all the interactions I have and all the people I meet.

25:29So when I typically go to an event, a large CEO event, my meetings and interactions would generate work for my team that would keep them busy for the next 10 days at least. Now they feed all of that. And before I get home, all of this is ready for the next step. And it is driving me harder. You know, I used to. What is that? Is that, for example, like following up on a potential client, you know, coming back on some different materials that you promised somebody? Is that the kind of... Right. Yeah. Let's say we have spoken to a CEO of a Fortune 500 company and I'd give my team a brief of, you know, how this company wants to transform being an autonomous enterprise.

26:13Our system will automatically generate a proposal based on the last public statement, everything available, the way their teams are organized, where the spans are and what it could be. And, you know, all of that is auto-generated. of course there's some refinement that a human being can add to but it's ready to go and how do you think about you know one of the things we talked about another episode was that it's kind of a new skill for some people to go from the idea of being a task orientated human to be more kind of somebody who productizes the the solutions into a bot flow and so that's what a product developer traditionally did.

26:55Now humans have to do that task. Do you have any kind of tricks on how to start to make that mind change so that you yourself maybe go, hey, I'm about to do these tasks over and over again. Can we please make an LM for that? I think that there is not an easy switch, but that happens over time. So I mentioned this task. Let's take this example of task. After I met a CEO of a Fortune 500, somebody was doing a series of tasks and it took 10 days. Now that all of it is available within first hour, it accelerated at an exponential pace the level of interactions and follow-ups and conversations. the nature of work fundamentally changed.

27:43I don't think even I was ready for it where after I met 10 CEOs, the next day I'm ready to have a detailed conversation with 10 CEOs. That's completely new, but it is exciting. That was the end result of it. Why should I wait for 10 days? So I don't think I fully adjusted to that, but I am learning to, right? That's just a new pace. Everything has a new pace to how fast you move. Well, the industrial, I think part of the whole industrial complex and task orientation, the reason that we lob an email back is because now we have plausibility to not do anything until we hear back from the person, right?

28:27And I've noticed that, you know, stuff that I used to work with people on, now I work with AI on. And the challenge is I don't get a break because it just gets back immediately. You know, I mean, like a silly example is I was taking a class like pre custom GPT is I was taking a class on how to build a bot. And I used to think, you know, uh, I post a question to the class discord and then the TA answers within 12 hours. So I post a question and I go to bed and I wake up, see if they got back to me. And if not, like, ah, I can't keep going on my bot. Right. Right. And the TA kept saying, have you asked you at GPT?

29:00Have you asked you at GPT? And I thought, dude, I paid thousands of dollars. This answer my question. I realized that he was actually teaching me the meta skill of you don't have to wait for me to respond. And when I realized that, I mean, that's when I started pulling all nighters because Chai GPT is a TA that responds instantaneously. And all of a sudden I realized, you know what I, and here's the point that I was making. I actually enjoyed quote unquote meeting to wait for another human because it gave me an excuse to stop working. And with AI, weirdly, we don't have that excuse of, I need to wait on this other human anymore.