Imagine agents, automations, and people brought together. The era of the autonomous enterprise. Instruments of work growing smarter, faster, and more capable. Multiplying what you can achieve exponentially. Streamlining workflows. Bridging silos. Harmonizing collaboration. Unleashing potential. Orchestration that defines the future. Imagine what's possible.
Please welcome the co-founder, CEO, and chairman of Automation Anywhere, Mihir Shukla.
Thank you. Thank you. Welcome to Imagine, everyone. To our customers, partners, we are honored you are here with us. And a special welcome to our new customers. We have accomplished incredible things together. Your usage of our agentic process automation platform is doubling every quarter. Imagine, every quarter. You are now running over half a billion AI agents in automation, generating immense value for your businesses. Last year we crossed $50 billion of collective value generated by all of you, and for a few of you, you are approaching a billion value creation goal. Congratulations. This is amazing.
This growth is accelerating as our offerings expand. We have expanded our R&D efforts by 50%, delivering more value to you in every product release across our APA platform, but also nine pre-built agentic solutions for faster results. These solutions aren't just transforming your business, they are transforming ours as well to serve you better. For example, we have applied agentic automation to our customer service, and today 50% of customer inquiries are resolved within 2 minutes. And we have used all the value created through agentic automation to hire 20% more people. Yes, 20% more people.
I think together what we are proving is that technology doesn't just drive efficiency, it expands opportunity. Through our partnership with you and social impact programs, we have reached over 800,000 people in the world. We have delivered over half a million hours of free training to help individuals welcome them to the digital economy, a future they may not have imagined before this. And because true innovation carries responsibility, let me say that again. True innovation carries responsibility. We are also deepening our commitment to sustainability. We are on track to deliver carbon neutrality scope one and two this quarter, way ahead of our plans.
There is a common theme in all of this: that agentic automation generates real business value. The efficiency it creates may lead to cost cutting in some areas, but by reinvesting that capital elsewhere, you can fuel tremendous growth. And none of this would be possible without all of you. So thank you. Give yourself a big round of applause.
Imagine is about you. Here you can learn from your peers, see many customer stories live in action, take a deep dive into our technology, and be first to see what is coming. Some of you might remember that about two years ago, we unveiled the industry's first AI agent. AI agent wasn't a term back then. And the industry's first dynamic process reasoning engine. And we told you that this is how the future will be. This year we're going to do something similar but take it to the next level. We will unveil the future of enterprise autonomy, and that will give you a view into what that future will look like eventually for all of us.
Now I know you can't go anywhere without hearing about AI. In the early stage of any technology, there is an explosion of ideas and a lot of noise, and that makes it harder to find your path forward. But I'm confident there is so much experience and intelligence in this room that we'll together find real clarity on this journey. Let me take a quick poll from the audience. How many of you use AI in your daily work life? How many of you? It's like almost everybody. That's wonderful. How many of you think that it is smarter and faster than asking the person next to you most of the time? That's why you use it. How many? Come on, be honest. You've been asking the person next to you in a cube a little less lately, right? It's getting smarter. We feel that. Yeah.
And it's not surprising because every decade or so we have a new definition of artificial general intelligence. We cross it and we make a new definition, as we should, because it is getting smarter and faster. People ask me whether our process intelligence model and reasoning engine model, how much smarter will it get? That's a meaningless question. How high can a plane fly higher than a human being can jump? It's not even on the same scale. In fact, that's not what we do. That's why we invented planes. This thing will be a thousand times smarter in coming years because nobody can train our models with five and a half billion input parameters, billions of runs. That's not what we were designed for. So that's a foregone conclusion that this thing is going to be a thousand times smarter, right? That's not what we designed.
So that leads to this paradox. If AI is this much smarter, why haven't we managed to transform many more enterprise processes? Why haven't we seen everybody achieve billion-dollar outcomes? What is missing? What is missing are three things that's slowing enterprise adoption down. The first is context. In enterprises, AI isn't about raw intelligence. It's about contextual intelligence that is based on your data and your business realities. You can't get a nuclear physicist to process an insurance claim, right? Can't. And if you just plug things together in a multi-step process without context, those errors expand and it is unmanageable.
The second problem we have is that we need a solution to this siloed thinking and the sprawl of AI everywhere. Because when you apply AI in silos, you relegate it to very specific task-oriented things, and it doesn't often deliver business outcomes. It doesn't increase revenue, manage costs, reduce risk. Think of it this way. Let's say you have 5% of the people in a company doing something and you did an AI that would improve their work by 5%. Collectively, that's a quarter of a percent for the whole company, and then there is a question whether it showed up anywhere. So we need to get out of the silos and to complete end-to-end business outcomes. We need to orchestrate between people and systems. We have operated in silos for too long. We're still operating in a post-World War II era where that's how since World War II we are organized. So that's what we are continuing.
And finally, we need governance and observability and security. They are the key blockers. Without it, no pilots can go to production. These are the three reasons why we haven't seen a significantly wider adoption. So how do we solve this paradox? What should be one of the most questions of all that I get asked is that what should be the enterprise AI strategy? Let's look at the options vendors are presenting. Many of you have played with co-pilots, and we all jumped into it because there is an appeal of faster value. We all want faster value, and it is great for personal productivity. But we now know for sure that it doesn't solve the paradox of enterprise productivity. It does not.
You could get AI from system of record vendors, and they are proficient in their specific domain, and we have trusted them with security and governance. So I can see that appeal, right? But the reality is that the system of record companies' incentives is the same old SaaS story where they charge for the seats and charge more money for a new AI technology. You hear from SaaS vendors that the revenue will increase significantly in the next few years, and it makes you wonder why would that be when 80% of your workload can be fully autonomous. Why would your revenue go up so much? Because they expect to continue to charge the old model while they build the new business model along with it. Look, AI gives us an opportunity to do things differently and unlock value. This is not against anybody, but this has to be in favor of value creation, and this is an opportunity of a lifetime.
There are also plenty of point solutions and AI startups, but the challenge there is enterprise-grade technology. You end up taking risk that you can't see. So with all these choices, what do we do? A collection of co-pilots, agents, and point solutions is not a strategy. And this is a technology of a lifetime. We think your strategy should be to become an autonomous enterprise. What is an autonomous enterprise? An autonomous enterprise is an AI-first organization where 80% of work runs autonomously or AI-assisted. 80%. You're reimagining work, and because agentic automation orchestrates across applications, it reduces the need for more software licenses. We need to decouple company's growth from SaaS tax. This model where the more you grow, the more toll booth you pay to every SaaS vendor doesn't have to be that way, right? It can be more outcome-driven. So that model has to change.
An autonomous enterprise does not mean no people. On the contrary, every employee should be 3x more productive. This is very important. Not 10%, 20% more productive. That would be another IT project. It's not an IT project. AI has to run an organization, and people have to be 3x more productive. Now, since we announced the autonomous enterprise vision more than two years ago, other companies have introduced their own versions. Agentic enterprise, agentic business. Some others have quoted now the same name even. But make no mistake, that's not the same. It has to have these three characteristics that allow you to reimagine how business operates.
Now let's hear it from an organization that is pursuing this vision. Today we are fortunate to be joined by leaders of the UK's National Health Service. Who doesn't know NHS? But NHS serves in the UK millions of people, and it is the fourth largest employer in the world with 1.5 million employees. Take a look at the video.
1.5 million employees, over 200 trusts. The UK's National Health Service is the largest single employer in Europe and one of the most extraordinary institutions. Like all large public systems, the NHS is navigating real financial and operational pressure.
Across the NHS, we're trying to transform patient care. I was having a conversation with someone who worked in the organization, and they were describing real struggles in the day-to-day processing of the organization. It became clear to me in that conversation that automation would unlock that person's potential and the potential of the organization.
University Hospitals of Leicester is one of the largest and busiest NHS trusts. 18,000 colleagues serving over a million people across three hospitals and seven community sites. It is choosing to lead by doing something the NHS has rarely seen: a chief people officer and a chief digital officer building a new operational model together.
We firmly believe that healthcare is a team sport. I could see a vision to create an agentic colleague which would free up the capacity of the people within the team to do the work that only they could do.
We really needed to understand the process end to end, and that's been the difference for us in this project.
The agentic process automation is making their working life better.
So much administrative work has increased over the years, which now automation should help us with and reduce that burden. So it's going back to the old art of being a doctor for the patient.
Working with emerging technologies, not only how great is the tech, but also how much do you trust the people that you're working with. And for us, AA was the perfect partner.
I think by 2028 I want to be walking around the wards, the clinical environments, listening to people, seeing the smiles on their face as they describe to me fundamentally how their jobs have changed.
This is what it looks like when a great institution decides to lead, and the team is here to show you how.
Please join me in welcoming from the University Hospital of Leicester CEO Richard, Chief Digital Information Officer Will, and Chief HR Officer Claire. Come on up here, guys.
We really have a unique opportunity to hear it from three different perspectives here. Thank you for joining. Hi. Hello. Thank you. You guys are very awake for this time. It's near midnight. Yeah. Richard, maybe we start with you. Give us the big picture. Look, this is impressive, but set a vision for us. What's the future of healthcare in the agentic era?
It's brilliant to be here, and thank you so much for the warm welcome in Dallas. I've been really excited about this conference for the last couple of months. Great opportunity to meet like-minded people and to learn. I think whether it is global healthcare, whether it's the provision of healthcare in North America, South America, Latin America, Africa, Asia, Europe, or the UK where we're based, I think the thing that brings us together is our three big priorities. So in the organization I work in, if you work in healthcare, I think they're similar. The first bit is around providing safe care to all patients, whether that's through curing them or preventing illness in the first place. The second bit is about being a great employer for all of your workforce. And the third bit is about delivering the money, working within our finances, being a more productive organization. There's different ways of delivering those three priorities. I think the best way of delivering all three together in a unified way is agentic AI.
Thank you. Will, how do you make this real?
Yeah, it's great because we've got a big vision about what the role of agentic AI can play in healthcare. And sometimes having a really big vision is terrifying, right? Because you recognize that you might fall short, but sometimes it gives you the permission that you need to be really ambitious and take some managed risks. And I guess that's where we are. So in terms of making it real, the first way is being really clear about what it is that we're collectively trying to achieve. The second part of that, I think, is really understanding the technology and the role that technology plays. I think for us we'll always need those foundational technologies like RPA and other aspects of automation. But what we've seen is if we're really going to deliver this vision around transforming the way we work fundamentally, we're going to have to move away from that traditional digital behavior of looking at a process and finding a way to digitize it and make it slightly easier, to fundamentally reimagining what that process is and what's actually needed in those scenarios. And for us the big shift has been towards agentic AI and agentic process automation because finally that allows us to have a system that can manage the context. It can manage the unstructured nature of lots of our work, and this for us has been the key enabler.
That's fantastic. It sounds like changes go beyond technology in this. Governance, teams, Claire, I'm sure human beings and workforce. Tell us a little more. What are all the other changes other than technology?
Yeah, so I think for us it's probably around three things that have made the shift and made the difference. So the first is that while the technology is fundamental, this is really about solving problems for our colleagues and for our patients. And so what that means in practice is that there are different people in the organization that lead on those processes. So Claire is a great example of somebody who leads on the people processes and really understands what the organization needs from that. So our first shift has been away from thinking about this as a technology project to thinking about it as solving problems in the organization where technology plays a supporting role. The second shift for us has been really about the return on investment. So, we've all heard big promises from tech in the past, right? That it's going to be completely revolutionary.
I don't know what you're talking about.
Yeah. And so we've been really ruthless in both tracking and realizing the benefits. So, we're starting to see a, we've committed to a 3.1 times return on investment. We're already starting to see that in our organization now just a few months.
Did you say 3x return on investment?
3.1. Okay. And then the final one really is about, and you probably would expect me to say this, but is about partnerships. Because lots of what we've done is stuff that you would expect. It's stuff that's relatively straightforward, but there's a huge amount of what we're trying to do which is really, really complicated and complex and hard. And there are times when you're looking for your partner, and when it gets hard, what you really want them to do is lean in.
Yeah. And for us, Automation Anywhere have been the perfect partner because when it's got hard and complicated and we've got stuck with a problem, it hasn't been like over to you guys. You know, you guys have really leaned in and given us some help and helped us understand this technology in a way we just never would have done on our own.
No, thank you. Thank you. Look, we've never helped the world's fourth largest employer before. So, we are learning as well. So, thank you. Claire, tell us, the huge side of this is human beings and the roles and responsibilities and how people think of work. Tell us, give us your perspective.
Yeah, sure. Hi everybody. So a few things about this, and you'll all be able to relate to this. Everybody can relate to healthcare. It's a business that is about people. It's about patients and it's about colleagues. And what the tech is doing for us is giving us the ability to totally transform the work that people do and how they do it. And for the first time, we're in a situation where the tech is actually working for the people rather than it actually being the other way around.
And what that's doing in practice for us in clinical care is, as Will and Richard have alluded to, we work in a complex environment. Our clinicians work in a complex acute environment, and what the tech does is it frees up their capacity. So it releases time to care. It gives them the opportunity to do only the thing that they can do, which is spend the specific time with the patient. And what it also does is release capacity so that within the same level of resources we've got more capacity for colleagues to be able to train and develop themselves. So it builds their careers, their portfolios, and their abilities, their clinical abilities, which means that they've got a greater opportunity to operate at the top end of their license. And the final thing that I would say is that it actually inspires hope and opportunity for change. So working in an environment where actually you can be constrained around resources, whether that's the availability of other colleagues, the availability of finances, or future investments. What this does is it gives people the opportunity to be able to see that change is possible. And I think that that's really inspiring. And where we get inspired colleagues and motivated colleagues, we know there's a direct correlation between improvements in patient care and patient outcomes. And they're the sorts of things that we're seeing. In fact,
I've not heard that before, but it's amazing. You know, so much of media narrative is not so much about hope, but it gives people hope, and that's a different perspective. I know we have a limited time. Maybe quickly, one message you would give to your audience, each one of you, what would you tell them from your experience?
Should I go first? Okay. So what I would say to you in the audience is that we've seen the ability of some of the tech. We know that there's more to come. The pace of change is fast. It, as I say, it gives us great opportunities. Continue to make sure that you absolutely focus on the people, build the team, and create the right conditions for all of the colleagues to come with you.
It's good advice, Richard.
I'm going to pick up on Claire's point about hope. So, we know that there are lots of challenges at the moment, but I also think there are many reasons to feel optimistic. So, I'm going to use a quote, and the quote is, 'Optimism is a duty. The future is open, the future is not predetermined.' And we can all shape the future with the actions that we take. I think we all have a responsibility for the success in the future, and I think agentic AI is going to be a key determinant of future success.
Spoken like a leader. Well, how do you follow that?
I guess my word would be to the technologists in the room, which is this isn't fundamentally about us. We get that collectively, right? Our job is to serve our organizations and to deliver the change for them. And I think increasingly as we move to mainstream this technology and it be about not about the technology but about how we work in our organizations, the role of the technologist then becomes one about serving the aims and ambitions of everybody else. And so I think for us we have to try and reimagine our roles not as people who drop technology into the organizations but as people whose job it is to enable brilliant people to do brilliant things.
That's great advice. Thank you so much for coming. Thank you. It's what a unique opportunity to get a perspective from three different angles. So much thank you. They are in a breakout, and I would encourage everybody to hear the story in more detail. It's an amazing story. NHS allowed us to see that the autonomous enterprise isn't just a vision. They are approaching their vision is to do 70% of everything that can be automated. Isn't that amazing?
Look, today I'm excited to share many new announcements that will accelerate your path to autonomous enterprise. And I warn you, there are many announcements, and during the next two hours you will hear more and more about it. First is our APA 2026 platform. Now everything that you love about it and the millions of automations and agents that you run on it is getting better. The latest release has an orchestrated, most orchestrated orchestrator that is now enhanced to deal with the most complex of the complex processes so that it can do big work that spans across systems with an accuracy that is required for regulated industries like in healthcare and NHS. Right? So here is the challenge. Bring it on. Bring it on your most complex processes on this platform. We can handle it.
The second announcement we would like to make is the 2026 version release of the autonomous ITSM solution. The autonomous ITSM solution is reimagining the economics of ITSM that is not based on tickets. The new version offers pre-built agents to resolve more complex types of tickets, connectivity to thousands of more applications. At last count, we have integrated 45,000 applications and a seamless user experience. It's a fantastic product. You'll learn more about it. And we are very committed to transforming the economics of ITSM. That world has to change.
The third announcement I'd like to make is in addition to enhancing seven existing agentic solutions with the next level of autonomy, we are also introducing two new amazing solutions: autonomous IT and autonomous finance. These solutions were designed by CIOs for CIOs and by CFOs for CFOs to automate an entire department. And I have seen it. It is unbelievable. You will see on the demo booths and during the keynotes later today.
Fourth announcement I'd like to make is we are making a new capability available called Automation Anywhere Code. Now, vibe coding is great, but the challenge has been how do you produce production-grade agentic applications with context, security, and governance. Automation Anywhere Code solves that problem. It helps you build, test, and deploy enterprise-grade applications with complete processes, agents, UIs, context, security, and all in as little as one week ready to go live. If we can accelerate that journey in a week, there are so many problems we can tackle. I know so many of you can't wait to get your hands on it, including our partners.
I know this is a lot of announcements, but I have one last announcement to make, and it is something special. I promised we'll show you the future of enterprise autonomy. Many of you have heard about open claw agents. Yes. How many of you heard of open claw agents? Quite a few. They can think, decide, and execute autonomously. The potential is real, but so is the problem. What makes open claw-like agents powerful for personal productivity is what makes them dangerous in enterprise settings. And I know you all know that because your most critical data lives in secure environments that were never designed to let unmanaged agents in.
Today I am excited to announce availability of Enterprise Claw. It brings together our enterprise-grade technology with the power of open claw-like agents. You will see other vendors trying to replicate this. But Enterprise Claw is different because we orchestrate and govern the entire process, the entire work, not just a few individual agents. The whole work has to be governed. And second, because it provides the accuracy that I talked about with the context that is ready for regulated industries.
Enterprise Claw unlocks unimaginable possibilities. For example, now you can take tickets like L3 and L4 tickets that are more complex, that often takes an engineer sometimes two weeks to diagnose what the problem is. Now, Enterprise Claw agents can quickly discover, sometimes within minutes or at the most an hour, what the underlying root cause is. That's unbelievable.
Now, we didn't build this alone. We built Enterprise Claw with partners who know what enterprise-grade means. The best in the industry: Cisco, Nvidia, Okta, and OpenAI. Cisco handles security monitoring and threat detection for agent activity. Nvidia provides the compute infrastructure and secure execution on-prem and edge. Okta, of course, manages identity and access control for agent interactions. And OpenAI brings the model and the model intelligence. And we have these partners here with us today. Please.
Hello. Thanks for having me this morning.
Hi there. Thanks for having me.
Yeah, it's a great question, Mihir. I think especially for this audience, I think they're more than aware of the kinds of issues that come up when you start having agents. The way I like to think about this is agents almost act like teenagers with a credit card, right? A lot of times they're extremely capable, but they lack any kind of judgment or fear of consequences or any kind of sense of boundaries. And so when you give an agent a task, especially an agent that has a large amount of reasoning capabilities, agents have a tendency to relentlessly pursue the goals of what it's being asked to go out and do, right? So, which means, at the end of the day when an agent's trying to be helpful, you really have to start to think about, hey, what kind of actions are the agents taking? So when you start to think about the threats, these superhuman characteristics of infinite memory, infinite patience, no fatigue, all of these things create an interesting threat model. And the way we think about this from a Cisco perspective is that that threat model is two-sided. You need to simultaneously protect these agents from the world because these agents can be attacked with attacks like prompt injection or context hijacking or malicious MCP servers. And on the flip side, you got to protect the world from these agents. You got to make sure that you're managing the access, you're managing your blast radius, you're making sure that there aren't any destructive actions being performed. So, you're fundamentally reimagining all of these capabilities, whether it's identity, whether it's access, whether it's behavior, not just at the task level, but even at that subtask level of these agents. You got to make sure that you're thinking deeply about these problems. So, it's a brand new world, you know, from a security perspective.
Listen, it's meant to unlock enterprises from adopting Enterprise Claw, right? I think that is the core thing. One of the first things that's going to happen to agents is security is going to fundamentally increase adoption. Security has historically been looked at as something that adds more friction, but this is going to be one of those cases where it's going to reduce the friction, allow you to use these tools a lot easier, a lot faster, right? And so with Defense Cloud, what you're getting is basically, we open-sourced it. It's a secure agentic runtime that directly hooks into Open Claw, hooks into Cloud Code, hooks into Codex, and those are the additional capabilities that we provided that gives you a skill scanner, it gives you an MCP scanner, it actually runs Code Guard, which means when you're using any kind of code generation tool, it looks at all of the code and says, hey, is this code appropriate, has it been, from a security perspective. So it brings all of these pieces together. In fact, since we launched, we've got north of about 3,000 plus GitHub stars, and some of the largest enterprises are now using Defense Cloud to be able to effectively secure their production environments. And the best thing about this is it really simplifies how you start to think about using these things from a security perspective, especially in large companies. It puts the CISO's mind at rest, and so that's really exciting.
Good to see you too.
Yeah. Well, I would say that AI is no longer a pilot. AI has become infrastructure. And a lot of customers, when they're moving from this experimentation phase to agents at scale, they need these agents to be governed, and they also need them to be very token-efficient because agents love to use tokens. And Nvidia provides that platform for deploying agents at scale and more securely. We have our NeMo platform for optimizing and tracing agents. We have OpenShell, which is our open-source secure runtime. We have our Neotron models for cost efficiency with models. And then we also have NeMo Claw. You were talking about claws. Claw is the big word. And we have NeMo Claw, which is our blueprint to be able to put all of these things together.
Yeah. So I would say that Neotron provides the intelligence as well as cost efficiency, and then OpenShell is again that secure runtime. It's that trust boundary because it operates at the infrastructure layer and a unified security credential layer that's separate from the sandbox to make sure that these agents, again, when you're working with agents, they have access to credentials, they have access to networks, they have access to files just like you said earlier in ways that we do, but we need to be able to give them policies and we need to be able to enforce those policies for them, and OpenShell allows you to do that.
Let me go to Eli. Eli, Okta of course is a leader in agentic identity, and you're bringing things like XAA to agents. Why is it important for companies to be thinking about this? Sorry if we switched to AI.
All right. Well, I think we all have to operate under the assumption that AI agents will be hacked.
I see prompt injection as a feature of AI agents, not a bug to be managed. It's really inherent to how LLMs process instructions. And because we can't patch this, the highest ROI security activity we can take on is securing the identity of AI agents and ensuring they have a small blast radius even if they are compromised.
What excites you most about us working together?
Well, ultimately it's all about creating least-privilege agents. The agents operating on endpoints are really the riskiest type of agents. They're agents that are susceptible to something called the lethal trifecta. They have deep privilege access, meaning they have like read and write permissions to your local file system. They have untrusted input processing. They're constantly reading external data like emails or Slack messages that could have malicious instructions embedded in them. And they can take autonomous action. They don't just talk, they can do. And because of this, we need to work together to ensure these agents have task-based fine-grain access controls. So they only have permission to do small tasks that they should be doing. And we need to ensure that the tokens that are created for these agents are ephemeral so that if they are compromised, these tokens aren't sitting around waiting for an attacker to use them to move across your workstation or ultimately across the company.
Thank you, Eli. Jason, let's just switch to you. Good to see you too, Jason. Jason, what is fueling this space of modern releases and innovation?
Yeah. And so you've heard from a couple people today how quickly things are accelerating. I actually want to take a moment to contextualize that. So OpenAI has had three model releases this year. We release on a 4 to 6 week cadence major model releases, and that used to be a multi-year effort. So if you look at GPT-3 to GPT-3.5, that was an 18-month training cycle. So if it sounds like things are getting faster, it feels like getting faster, they very much are, purely in terms of the pace of launches. And it's also true with the cost of intelligence. So the comparable sort of AI capabilities today versus 18 months ago, the cost of that has come down on the order of 100x. So you're seeing much smarter models but also much more efficient models, meaning you can deploy them in more places. And so to your original question in terms of why that is, there's so many resources now in this AI industry with the best people, smartest people in the world working in it, that we've seen improvements across the stack. That means hardware, that's pre-training, post-training, and all of those combinations of improvements have led to this continually accelerating cycle that we see.
Yeah, Jason, what are the biggest opportunities for our joint customers as part of working together on a project like Enterprise Claw?
Yeah, so as a byproduct of these increasing model capabilities, the agent capabilities are not static. I talked about how quickly things are accelerating. What we're seeing is that there's a big overhang in terms of what the models are capable of and then what you actually get out of it within an enterprise. And the biggest reason is because there's a gap between those capabilities and then also the governance, the security, and the observability required to be successful in deploying that. And so we're very excited about this partnership. Automation Anywhere has 20 years, over 20 years of enterprise process automation experience. You need the right agentic infrastructure in order to secure and make sure these deployments are successful.
Thank you, Jason. Appreciate all of you joining us today. Please give them a big round of applause. Together we are imagining the future. Thank you. Thank you very much.
Enterprise Claw is this incredible capability area. Take an example I mentioned about L3, L4 tickets. Think about taxes and mortgages and claims and supply chain. In each of these areas there are enterprise claws, and there are a few of our partners here who are announcing solutions on it that could create a thousand people's worth of work in one hour. That kind of capability, exponential power, it's incredible what is possible. Look, in addition to the industry partners, several customers have also helped us shape our thinking for Enterprise Claw, and one of those is Citi Bank, which is leading a bold vision to reimagine banking.
Please join me in welcoming Citi's head of functions technology and head of enterprise change, Julian. Julian, come on up.
Good to see you again. Thank you. Thanks for coming.
Julian, everyone knows Citi, but maybe just tell us the size and scope of it. I think this picture does it justice, but still it's beautiful.
Yeah, I mean, again, thank you so much for having me here. We have a great partnership, and I appreciate the opportunity to speak to everyone here. So I've been at Citi for under two years, but I will tell you every day I'm amazed at the scale, scope, and complexity of the organization. We serve clients in 90 countries. We have 280,000 people and employees. In the technology organization which I lead, we have almost 90,000 people, 45,000 developers.
Unbelievable. You have more developers than many software companies. The scale is fascinating.
So tell us, how are you architecting Citi for the agentic era?
Yeah. So we've been really looking at AI or GenAI since really 2024 where we've started to first do our first implementations. We've done AI of course like any other financial institutions for many years in our trading and market environment. But really looking at GenAI since 2024, we work with most of the large frontier models. We've developed an infrastructure that enables us to connect our frontier models with our applications with MCP services and what have you. And then we have a lot of different point solutions across the organization, and quickly we realized this is just not going to scale, right? What our differentiator is the scale of our institution, the scale of our ability to serve clients. So we needed to develop one orchestration architecture, one development framework where we would put all of our AI capabilities and then integrate across our businesses. And so we have actually developed an agent development framework environment. We're looking at all of our core processes across Citi, and then for each of those we're looking at how do we actually orchestrate our solutions because we have so many developers, we like to build our own solutions, but we also realize that for speed and efficiency it's good to partner with really strong capabilities in the market. That's where we have a lot of good relationships with you and at Automation Anywhere.
And specifically in the orchestration layer space, right? How do you define the success criteria? Accuracy in a regulated industry, very important. How do you decide?
Yeah. So for us there are really two components. One is client service, right? How do we actually bring AI to better serve our clients, to bring better products, to accelerate how fast we can actually deliver services to them? How fast we can open an account for them? How fast we can approve a loan or what have you. So really it's about speed. And then two, it's of course about accuracy. And so I think you mentioned earlier the concept of the semantic layer, and that really brings the context for the agents, and that's critical for us, right? We have probabilistic solutions, and we need deterministic answers, right? Deterministic responses to problems. And that's where we actually build that semantic layer where we can train the agents to actually learn the business, learn the clients, learn the data, and then enable them to make sure that the answer is systematically the same when the same question is asked.
It's something that resonates with every regulatory industry. How do you take probabilistic agents but we need deterministic answers, and how do you find that balance?
How do you, many of the companies in this space are stuck in pilots, and some people say there are more pilots than airlines. So how does Citi of your size not get stuck in pilot?
Yeah, so it's a really, really good question, and I'm sure many people in the audience feel the same, right? Where you have a lot of different AI-based solutions that do not scale, right? And if it doesn't scale, it doesn't really bring the value that we're looking for. So we actually, as I mentioned, we started with having a lot of different pilots and agents kind of in different parts of the organization, and what we did is we decided to actually centralize all of it. I know many of my peers are saying they have 400, 4,000 use cases. We actually decided to focus on 55 use cases.
55 processes. That's our wave one. But we wanted end-to-end processes, right? We wanted to start with the customer and then end with whether it's the processing of the decision and with our reporting capabilities and with our finance and risk environment or back end if you will. And for each of these processes, we actually deploy this orchestration layers. We actually deploy the agent solutions. We started also to look at our risk and compliance environment. And we decided to identify risk and compliance as well. So build agents that would enable risk managers to accelerate their review of the pilots and the solutions and the capabilities we develop so that we could go a lot faster in actually shipping products to the market and developing new solutions for our clients.
That sounds like great advice for all regulated industries because one of the biggest hurdles is compliance and regulations and risk awareness. But if you help AI there, you can accelerate the entire AI journey.
Exactly. Right. And we want to demonstrate that velocity doesn't go against governance. Right. So we bring the two together, keeping human in the loop, making sure that risk managers are reviewing the decisions, reviewing the output of agents, and then providing a decision.
