What does the future of software look like in the age of AI and agents? Let's talk about it with Amit Zavery, the president, COO, and chief product officer of ServiceNow, who's here with us in studio in a conversation brought to you by ServiceNow. Amit, welcome.
Thanks for having me, Alex.
So, this is the beginning of a series of four conversations we're going to have with ServiceNow that's really going to tell our audience how a software company handles this moment of AI. And it is a fascinating moment that you are in. Especially, you know, the question is, do you harness the AI? Do you fight it? Will you be displaced? Will you be able to serve your customers in brand new ways? And I think as viewers go through this series, they're really going to understand how a company like ServiceNow attacks this. And I want to give folks — for folks who have loose familiarity with ServiceNow or don't know it very well — just a little bit of a description of what the company is.
Well, sure. Let me give you a background on ServiceNow. ServiceNow started 22 years ago to automate business processes and allow all the employees to get much more productive and efficient. So it's building a platform to allow companies to really work hard through a software evolution and make their business and the processes much more efficient. Yes, that's what we continue to do.
Right. So, ServiceNow handles IT operations like IT tickets, customer service requests coming into the company, HR requests. This is just the software backbone that consolidates all of these and centralizes all of them. Market cap is 90 billion plus. 90% of Fortune 500 companies use ServiceNow and you handle 100 billion workflows. Very big company, really in the thick of things. And I think let's just talk about the elephant in the room here. We are in a moment where there is a narrative that AI is going to eat software companies like yours, where we're in this sort of — it's called the SaaS apocalypse — where, you know, why is there a software company providing these services? AI will just do it on its own, and then therefore, you know, either we don't need these companies or the customers will be able to come back and negotiate better terms because the AI can do something otherwise.
Yeah. I think there's a lot of noise out there, no doubt, right? And people are not able to understand what AI does and how it helps you and where it's going to impact you positively or negatively. So it creates a lot of confusion in everybody's mind — buyers, customers, users, analysts, investors, and everything else. So I can understand that unawareness creating that anxiety level higher. The reality is a software company like ours — and we have been investing in AI for quite a few years already. So it's not like AI suddenly we woke up and said there's AI. Our products and technology have been evolving over the last many years to be AI-first and AI-native. So we're using AI as a tailwind to really improve how we deliver new technologies to our customers while keeping the safeguards and the guardrails and things like that, so customers don't end up having issues around security or compliance or audit and things like that. Right? So today enterprise customers are very different than consumers. And what happens when people talk about AI today — you know, in the consumer world you can just go to a different website and say I'm using somebody else's product. Enterprises don't work that way. There are hundreds of years of technology which they've built out over time. They've taken a lot of new products and invested into it and integrated into the existing environments, and they need the business to run efficiently and properly without having errors. AI doesn't guarantee you answers. I mean, I think you had Mark Cuban last week, I believe, on your podcast, and he just had a statement on X yesterday talking about how every question you ask AI, you get different answers.
So imagine if you are doing your financial reports and if you get a different number every time, would any investor believe your results? So just taking AI by itself is not going to solve the problem. You have to bring AI with the guardrails, the harness, the enterprise domain understanding, how everything connects together, how it works, to really make it much more efficient and usable. So bringing the probabilistic nature of AI to the deterministic nature of workflows — what we've been building for many years — and bringing those two things together can be a game changer. And then how you monitor and manage it, because the AI agents which you're talking about nowadays are changing their identity, their access, their ability to do things every few seconds. So you need a lot of things around something we call AI Control Tower to really have visibility, control, observability, and cost control around it, so that you have a full idea of what's going on. So a lot of these technologies are required. So people, I think, are misunderstanding that AI will take out all the software companies. I'm sure there's some software companies — just like with any transformation happening in technology — will have a problem, right? Which happened during cloud, which happened during web, and that will happen with AI. But companies who are going to harness the value of the new technology and bring it into their products can really accelerate the business. Look at our numbers. Our business is growing at 20-plus%. Our free cash flow margins are at 35%. Our operating margin at 32%. And we have guided every time, beat and raised our guidance and consensus in the market. So looking at results, the value you create, and the customer validation — at this event we have 22,000 people here, more than what we had last year. So there's a lot of demand there. So I think there is a narrative, but you have to differentiate yourself from what we can do with it. Can we really add value to customers, and can we prove it? And that's why we will be the winners. But there will be some losers out there. Sure, why not?
Yeah. And we're here at Knowledge 26, which is ServiceNow's flagship event in Las Vegas. So this is where we're going to be conducting these conversations. Let me debate you on this a little bit. So ServiceNow handles a lot of IT operations. The argument that some would make would be, all right, well, you know, why will I have to route a conversation into an IT desk if, for instance, my agent can take control of my computer and fix my problems? So the idea would be that the function actually is transmitted over to the agent and therefore doesn't have to come through a platform like ServiceNow the way it has in the past.
Yeah. First thing, you remember the systems you are touching at the back end — probably hundreds. Most of the enterprise companies have 300-plus systems. When an IT request comes in, first we already do this thing where you self-select them, right? So you can get self-service capabilities using AI. So in our products today, something we call Employee Works, you can ask a question in the environment and say, 'I'm wanting some help associated with an IT issue or HR issue,' and we'll figure out what your issue is and then route it accordingly to the right kind of systems in the back. What we introduced this week is something we call AI Specialists, who are not taking over the full human work, but doing it with the full security, right kind of privileges — not wiping out the systems for you, but really making sure we solve your problem in a very short amount of time, but with a lot of the understanding we've had for years. There's something called context. The context is important because you have to not just say I want to do something — why you made a decision, who made the decision, what were the guardrails, as well as all historical information. You can't learn that in five seconds. You have to do that over years, and we built that for years. So our ability to fix a problem when somebody needs help is much faster, much cheaper, much better, and a guaranteed outcome. Because with AI by itself, the agents doing it themselves — we also use agents, but we put a lot of harness around it. We put the controls around it. So my outcome with the context engine is going to be a much better outcome, where the customer says, 'You know what, I got my answer. I don't have to repeat it again.' Many times I've seen AI by itself hit it only 30% of the time, 40% of the time — the 60% of the time it fails, then somebody has to go and do it again and again and again, and there's no guarantee that it'll be right, or you might do something wrong or nefarious and go make changes to the system which you're not allowed to. You saw what happened with Ticketmaster.
What happened? Tell our viewers.
Ticketmaster — there's a large travel agency. What they were using Cursor and technologies with AI agents out there to go and maintain the codebase, and the system basically wiped out the customer database and the production system in nine seconds. Now the CEOs out there — you can talk to them about the issues they're running into where they have to recreate the whole system again. They don't know who booked what, when are they supposed to support them, and there are a lot of examples like that. So you need harness, you need guardrails, you need context. AI is very powerful, very valuable. But just saying that I'll run some agents myself as a desktop and I don't know what they'll end up doing — they have no control. You don't guarantee results. You will have a problem, and enterprises cannot afford it with the millions of dollars of losses. And what is the value of doing it? There's software which works. We do this with AI. You're replacing something which is working with something which doesn't guarantee you anything. And then who will maintain it? See, software is not just the idea that I'll run it once and I don't need to evolve it. So who's maintaining the software for you? Who's keeping up with it for you? Who's guaranteeing the results? And when something goes wrong, who are you going to call? Your key people — most of the people have domain knowledge about your business. They don't need to be worrying about running systems and building things and running it, because that's not the job. Now if you're taking them away from their day job and doing all these things for saving probably nothing, it doesn't get you anything, right? And we've seen that — we've done numbers, calculations. The cost of building everything yourself with all this technology, which is changing every five seconds, is 5x to 10x of what you would buy from us. So once you start doing the ROI, the TCO, the numbers don't add up for the customers, and people are missing those kind of things because they're just thinking that, you know what, I can go to a different browser link in my consumer world and apply it to enterprises. I've been in the enterprise business for 30-plus years, and we've gone through multiple technology shifts. Each one has its winners and losers, no doubt, but you also have to be thoughtful about how you approach it as an enterprise customer. You can't just rip and replace everything, and you have to see the benefit for it — what are you getting out of it by saying I'm going to have agents do this? If there's no ability — most CIOs do not allow you to do it. You don't want to have some random user go and update things inside a system without permission, and that agent might end up doing something wrong as well. What happened with Ticketmaster — if you ask the question, you should read the talk track. They asked the agent, 'Why did you do this?'
Yeah. 'No, no, I know I'm not supposed to do it, but I did it.'
That's it. Like, what are you going to do after that? You going to fire the agent? Nothing you can do. You're stuck.
Right. And your business has gone through a problem.
Those are many examples, and Ticketmaster is one of them, right? It's not a question that AI is useful, and we use agents, but we put a lot of work on top of it to make it work for us enterprise customers. This AI Specialist — so we have this idea of a level-one support engineer. It's completely AI-driven, right? It can replace everything humans do, and we can close a case or resolve an issue in 20 minutes versus two hours. But that is guaranteeing you a result. So my cases are going to be resolved 90% to 100% of the time. Humans resolve 60% to 70% of the time. But I have done a lot of context in it. I'm not just giving you an AI agent. I'm giving you a full solution. And for a customer, if you talk to any customers today, they're like, 'I want a solution. I'm not trying to do spare-part work. I don't want to be a rip-and-replace team. I don't want to build everything from scratch.' And if you look at the models nowadays, right, it changes every few weeks.
Right. But the argument is that the change isn't that it becomes different — it gets better. And so this is where — and as you're giving this answer, I'm thinking, well, this obviously makes sense. You have domain-specific information. You're working in secure workflows. You're doing things that you'd be foolish to trust to a Claude Code. The argument — and this is sort of where I'm like, so how could anybody think that this SaaS apocalypse makes sense? And by and large, I don't think it makes sense. And I've been on the record about this. But if I'm trying to get into the minds of those that believe it, what they see is this exponential technology growth where they're not sure where it will end. Maybe it will go towards AGI or superintelligence. And then if it does, the entirety of software looks different, doesn't it?
Yeah. The software is changing. So look at the way we build software and the way we deliver. We're using, harnessing the power of AI. So it's not like we are ignoring the value AI brings, but you're combining it with a lot of other supporting structures to make sure the business is safe. So AI, as you said, is moving fast. But if you say, 'You know what, I will build everything myself,' I have to keep up with it individually. So my enterprise user having to deal with it — like, I'm going to build everything. Right, build was always existing. Build was not like something new came up, right? We always competed with build versus buy. People could buy build software, they could build. Always in ITSM, there's nothing like you can't buy build with software. We build with software too, right? So anybody can do it. It's the ecosystem around it, the connectivity, the testing around it. So when you do this — are they crazy in thinking that agents can do all that stuff then? They are by itself 100%, right? Because what you'll land up doing is you have to keep updating. So your prompts will have to change again when the new version comes out. You have to do a lot of testing again. You have to make sure that it's kind of doing everything you expected. But the AI can't prompt and test and — no, it doesn't. I mean, the backward compatibility is missing because the environments are so many, so varied. I mean, I said this is not like small systems. If you look at companies — I mean, you look at large enterprises, the Fortune 500s. If you look at IT systems there, it's not easy to go and just rip and replace all these things and then say suddenly introduce something which doesn't prove itself, and then keeping up with it and the backward compatibility — all this stuff becomes very... People are underestimating the security part of it. The compliance — look at the amount of time engineering products like ours we spend on compliance. 32%, 40% of my cost structure is compliance with different regulations. There are hundreds of new regulations coming out. Who will do that inside a company when I buy an LLM? LLM is — 'Yeah, I can, yeah, sure, I am compliant.' Does it guarantee you that? I don't know. Does it give you all the different support structures around it? Does it pass all the audits? Does it — when something goes wrong, can you now know how to return to the normality before? They would just move on. So I think there's nobody to really manage that for you, and then say, 'Okay, now I'll have to hire a lot of engineers to maintain it.' Then when you add up the cost again, it goes up beyond what I can provide you, because we have people who are domain experts who've been building it for years. I can scale my business without adding incremental cost every time a customer buys from me. But now every customer is going to replace and replicate what I do — it's going to be costly for them. It just doesn't make sense. So if you talk to any CIO — and I think if you saw the keynote today at Knowledge, you had Vishal and Raj from the CEO of FedEx there. They were talking about, like, even if I could do it, why would I want to do it? Vishal's point is that he's the CIO of FedEx, one of the largest logistics and supply chain companies there. They said, 'I have other things to worry about — things which work and I know are proven — and they're bringing AI into it. Why would I want to replace something which is giving me safety, risk management, and guaranteeing me the right results? I'll use AI for things where I have to build my own proprietary stuff, but things which are proven and working well — you partner with the companies who can do that for you.'
Yeah. And if you try to rip it out and build your own and it doesn't work, or you create a Ticketmaster situation.
