Copying One ServiceNow App with a Language Model Costs Ten Times as Much

Bill McDermott opens with a spreadsheet, not a speech. Replacing an enterprise platform in the “SaaS apocalypse” people talk about is, he says, an extraordinary expense: the rebuild itself, the human capital diverted from work the platform was already doing, and the GPU factories and tokens that will “materially affect” a language-model company’s business model. For a simple application on ServiceNow’s platform, that stack comes out ten times more expensive than keeping the platform and adding a model. People who run businesses, he adds, will forgive people who make mistakes. They will never forgive software.

Sarah Guo introduces him as ServiceNow’s CEO and “the closest thing we have to a rockstar in enterprise technology.” The conversation she wants is about leadership in the age of AI, what enterprise customers actually ask for, the gap between SaaS platforms and SaaS-apocalypse theory, and the next decade of the company. McDermott, who also ran SAP, has the unusual résumé of someone who bought a deli as a teenager and now sits in the corner office.

Guo had just finished his memoir *Winners Dream*, and she wants to start at the beginning. The book’s arc, from a Long Island corner store to a global software company, is the same argument he will make about AI: technology is supposed to serve people, not replace the ambition that got them the shot in the first place. The deli is not nostalgia. It is the operating system he still uses.

At sixteen he was stocking shelves, pumping gas, and bussing tables. A neighborhood deli came up for sale because the owner was going through a divorce, and McDermott treated the purchase as a way to collapse three part-time jobs into one. He bought it for a $5,500 note, $7,000 with interest: make the payments and keep the store; miss one and they take everything. He does not claim genius. He claims he got lucky, and that the luck came with a brutal rule.

The rule was the customer. He had three of them. Blue-collar workers like his father, “rich on Friday and dead broke by Sunday morning.” Senior citizens who never wanted to leave the house, so he delivered, “the early days of DoorDash.” And kids he had to pull a block and a half past 7-Eleven, which he did with Asteroids and Pac-Man cabinets that ate their parents’ quarters for hours. At the end of a long day, one of those kids told him the quiet part: they came to his store for good food, dignity, and video games. “When we want to steal stuff, we go to 7-Eleven.”

The emotional quotient of 500 customers a day, he says, is still worth a lot, and the world could use more of it. Guo recognizes the drill: she was a hostess at Outback Steakhouse, and the repetition of dealing with many kinds of people is its own education. Know who the customer is, give them what they want the way they want it, and you have a chance in any business. The store put him through high school and college. It also gave him his start.

He had worked there first, which is how he saw the customers before he owned them. The lease was only a year, held by Sun Mark Industries, the Sunoco gas-station landlord, which could throw him out anytime. Only a kid who did not know better would pay for that, and he had no cash, so he took the note. What he did have were supplier relationships. He talked them into the first order on consignment: fill the shelves, I will sell your goods, and I will always owe you that first delivery. They probably overcharged him on margin. He did not care. He was in business, and all he has ever wanted, he says, is a shot. An opportunity is a gift. Cherish it, and make the most of it.

The night before the taping he had been with about 800 business people in from Brazil, talking about leadership and about *Winners Dream*. Leadership, he tells Guo, is the greatest profession in the world, and the environment now is faster and more dynamic than it has ever been. His line for them was not comforting: it is fast, “but it’ll never move this slow again.” That is stressful, Guo says. It is also inspiring, he answers. Cope with the challenge, especially overcome it, and you get a superpower.

He keeps going back to the first break: Xerox, “the Google of its era.” Newly 21, his father dropped him at the Long Island railroad, and McDermott guaranteed he would come home with an employee badge in his pocket. Dad told him not to put that much pressure on himself. On the train he read the annual report on CEO David Kearns reinventing the company with Deming’s quality principles, at a moment when Japanese firms were selling higher-quality products for less than Xerox charged for worse ones. By the time he reached Manhattan he wanted to be the next Kearns. He would still have to knock on cold doors to get the shot.

The last interview was at 9 West 57th, 38th floor, overlooking Central Park, with Emerson Fullwood. McDermott decided it was not an interview. It was a battle for survival: get the job and you control your destiny. When Fullwood closed with the usual “HR will be in touch,” McDermott told him he had not broken a promise to his father in 21 years and had guaranteed the badge that night. Fullwood tilted his head. “As long as you haven’t committed any crimes, you’re hired.” McDermott hugged him, put him safely back in the chair, ran to a payphone, and told his parents to break out the Korbel, the cheapest champagne on Long Island, which where he came from was Cristal.

Twenty-five years later Fullwood asked him to speak at Rochester Institute of Technology and, introducing him, said it was the only time he had ever broken Xerox hiring policy. McDermott had not known that. The lesson he took is that you have to put tremendous care into people, and into what unique human beings are capable of. Fullwood still comes to his CEO meetings. Agency, he tells Guo, can be taught: he was a shy, quiet kid who came into his own through work, knowing he was not making the NBA. What he could control was how hard he worked and the desire to make a difference. Simulate, train, certify, coach, and put people in front of other people. AI should serve people and make human ambition greater, “not to take it away from us.”

Guo notes that his first reaction to the scary pace of AI was that it is also inspiring, which is a lot to ask of someone running a large platform company. McDermott does not deny the fear. AI has disrupted every industry, every business, and every person, seemingly all at once. His response was to lean in. ServiceNow wrote a white paper, a blueprint for agentic business, whose first job was to mark the boundary between a language model and an enterprise platform. Not better or worse: both matter, and each makes the other stronger. CEOs have told him the paper let them “see the world totally different.”

The worked example is a compensation dispute on aisle seven. The language model does a great job: in a situation like this, consider steps one, two, and three. It does that in milliseconds. The only problem is it does not close the case. If the person on aisle seven is in sales, HR is involved; because it might be a problem, finance, legal, compliance, or risk probably are too. Data and a workflow have to move through several departments, databases, and iterations before the issue is remediated and the case is actually shut. That, he says, is what a workflow platform like ServiceNow does.

The unlock he wants memorized is “AI think. Workflow acts.” The other case is a bank with 15,000 of its own engineers, a CEO under board pressure, and a meeting in 60 days that needs an AI story. New code from models is beautiful and faster than anything the industry has seen. What the apocalypse narrative skips is the cost of matching what a serious enterprise platform already does. You would have to rebuild it, keep people on that rebuild instead of other work, and then pay for the GPU factory and the tokens. ServiceNow has done the math. For a simple application on its platform, replication with a language model is ten times the cost.

Even if you accept the model, it is probably right and still not deterministic, and it does not have decades of company data and relationship context. “People that run businesses understand that people make mistakes. They never will forgive software for making a mistake.” At a small fraction of an IT budget, a premier platform is a bargain; combine it with AI and you get innovation, integration, and an outcome, plus someone to call when it breaks. Guo adds what the apocalypse talk ignores: most businesses do not want to build workflow software for everything they do internally. They want someone responsible for it. Gathering that context still takes a lot of energy. It is not automated yet.

Long-horizon agents and computer use do not change his mind. Language-model companies are fantastic and will do fabulously well, he says, the same way he once talked about hyperscalers. It was not long ago that people asked why those clouds would not just eat software, because every workload would live there. ServiceNow’s answer was to become the AI control tower for business reinvention: integrate with all the hyperscalers, all the language models, and all the systems of record, and be the fabric that connects those nodes so a customer can run a true agentic business. On the back of the ServiceNow franchise, he notes, billions of dollars of workloads already go into those clouds. They do different things super well. Choice, including choice of models, is the point. Staying out of the way, while acting as the tower, is the differentiation.

Security is the other move, not because ServiceNow will do all things in security, but because of a ranking he likes to give: the United States is the world’s largest economy, China is number two, and cybercrime is number three. It is “a $1 trillion a month problem.” With Armis now inside the company, he wants corporations to see IT and OT together. Operating technology is the rest of the landscape AI may not even see: devices on a network, manufacturing tools that may not be connected, medical devices in a hospital, a bank’s critical infrastructure, the building and the things in it. One automated AI platform with a full purview, he argues, is a compelling pitch for reinventing the business.

The platform itself is now autonomous, he says, and ServiceNow is a native AI company. Moveworks is the agentic front door to the enterprise. Veza covers human and non-human identity. Armis is the security layer on top of the core. The detail he is proudest of is speed: they integrated those businesses in 20 days. “A lot of companies don’t have the engineering power to say they can do hard things quickly.” Guo has sold companies to large enterprise vendors where it took a year and a half of rebuild to get any value. The historical rhyme is the one she wants remembered. Hyperscaler fear became partnership. Workflow software did not go away.

The installed base is the other half of the moat. The last time he looked, more than 85 billion workflows were in flight while they were recording, with 7 trillion transactions happening on the platform, representing some of the most important brands in the world. That is pride, he says, and a serious responsibility. Those companies cannot miss. The system always has to be there. Fortunately, he adds, it is.

Guo asks the sharp question: is any segment of enterprise software actually at risk from generated code and agents? McDermott says yes, some companies will be, even though data and context around that data should theoretically make them more valuable. The risk, if there is one, is departmental. Firms that span multiple departments, or that are a serious system of record, are very hard to replicate and will be part of the future. Serve one function, with work that is not tremendously high value-add and may not even be high on a CEO’s list, and “the vulnerability equation really rises quickly.” ServiceNow’s luck, as he tells it, is going end-to-end on the horizontal and north to south on the vertical, up and down the stack and across the enterprise.

He has run two platform businesses, SAP and ServiceNow, and Guo wants a definition. If ServiceNow had stayed an IT service desk, it might have been a system of record for the IT department. CEOs look for needle movers. SAP, in his telling, runs the financial system of global corporations across currencies, geographies, and industries, mass-customized, with huge switching costs; that is a platform. ServiceNow is a system of record for IT and the services around it, but bigger in another sense: roughly 800 of the more substantial systems of record integrate into it. Underneath is an in-memory open-source data platform that will ingest as much as you give it. If you do not want the data in ServiceNow, they make a zero copy: the transaction completes, nothing moved, no extra risk. You can change a business process on the spot with AI tools already in the platform, including seeing the impact in a knowledge graph in real time. That motion, he says, is very hard to replicate.

Implementation cost used to be the tax on all of this. Guo asks whether AI is bringing it down. McDermott calls it a big advantage. Nobody has time for multi-year projects, and with an autonomous platform, major customers are going live in less than 30 days, in some cases faster, in modular chunks so they can take quick wins and keep going. In the financial-crisis era, even SaaS took months; a 30-day live date would have been a rain dance. Some people thought systems integrators would be hurt. He sees the opposite: more projects, done much more quickly, with better customer ROI. Everything ServiceNow changes now, he says, is because of AI, including how humans actually work inside the company.

The company is its own lab. Ninety percent of ServiceNow’s customer-service cases are now managed by agents; only 10 percent still involve people. Work is lifting and shifting from grinding out details to critical thinking and judgment calls. McDermott thinks that makes people happier in the work, and he does not pretend it leaves the work unchanged. Fast-forward five years and he expects net new headcount to be dramatically reduced, because the company will be far more productive and the agents are real. Finance, HR, and supporting functions used to need thousands of hires just to keep up with a growth company. Agents will take a lot of that. Investment goes to humans who engineer innovations and humans who manage the relationship: trust, promises kept, the net present value of loyalty. That stays human.

The bar on skill, he says, just went up, at ServiceNow and everywhere else. If an agent can do the job as well or better, that is an easy economic decision. It is why he expects 2.2 billion agents to enter the workforce in the next couple of years, a lot more agents than people. Guo puts the company’s own scale on the table: about 20 percent annual growth, more than $13 billion in trailing revenue, a massive business still compounding. The SaaS-apocalypse narrative sits oddly next to those fundamentals. What he looks for internally is whether the AI control tower is actually being adopted: agentic front door, human and non-human identity, a workflow data fabric that all the nodes run through. CRM is the fastest-growing TAM. He also watches the trend line of agent assists to humans in customer companies. If consumption of that innovation is going “right and up,” the platform has solidified itself as how the company actually runs, and the consumption model gives shareholder value a different meaning.

The customer base is not where the keynote is. Most people know they have to do something with AI. In that Brazil room, only 11 percent of the companies had gotten past the experiment phase. Public sector is still consolidating complexity so one department can speak to another and billions in fraud stop falling through the cracks. Healthcare is in a modernization phase, because the professionals are great and the systems have been holding them back. Financial services is moving at full speed, rethinking how the company runs and how to get more done with less. Geography matters too. The United States is the leader in AI consumption and adoption. First movers are going for it; headcount is on the table; new people get added only with a clear ROI case that those heads support digital transformation and the agents.

Almost nobody, he says, talks about 2020 to 2024. Companies hired a lot through COVID, much of it virtual, under a keep-the-people logic, and layered up. Some of those hires never got a close look because Zoom was how you met them, greeted them, and hired them. Now, with AI in the business model, the question is how to get leaner and smarter. If AI really makes the company better, it should show up in profitability, revenue per employee, expanding margins, and reaccelerating revenue, a more exciting growth story than two to five years ago. Expectations for what AI can do are high. They are also, finally, numerical.

The biggest change in customer conversations over the past year, McDermott says, is a single demand: “tell me what I need to know.” It is no longer let me come in and discover your problems, or let me share the solution I happen to have in my pocket. If you do not know the business, there is no conversation. If you do, be prescriptive. At the macro level, how do we transform in keeping with the CEO’s vision? At the departmental level, understand sales or human capital or engineering or the CTO, and bring solutions that get up and running fast, deliver value predictably, and are AI-driven, because customers know AI is the force multiplier. “The dance has gotten brief.” Execution is where they want to spend. Get me there fast. They are impatient, Guo says. Decisive and fair, he answers, and open-minded if you can actually help, because they will tell you they need the help. They are not pretending they have all the answers. If two people are in the same room at the same time with the same opinion, he likes to say, one of them is redundant.

Outside ServiceNow he is curious about how technology will actually improve the human condition: environment, outer space, a global company simulating a presence in a market it has not physically conquered, new routes and channels that were not possible three to five years ago. He goes back to the Robert Kennedy line he put in the book. Some men see things as they are and say why. “I dream things that never were and say, why not?” That is the posture he wants for the era, looking at what does not exist yet and asking what technology, or a great dream, makes possible.

A day in the life is a rotation through time zones: start with Europe, migrate to the United States, wrap with Asia. He tries not to make every decision, but he offers observations, riffs honestly, and then they have to get someplace quick. Grounding still comes from customers, because the customer alone determines whether you win or lose. The other habit is employees. The day before the taping he started with 17 conversations with 17 quota-carrying reps; he has 72 on the calendar that month. Anytime you need to feel good, he says, talk to the people closest to the customer. He wrapped that same day with the 800 Brazilian executives. They are uncertain, even afraid. His advice was not to look for the escape hatch. As the waters get choppy you see who is tough, and when the tide goes out, you want to be fully dressed. Even if things do not look good, you will figure it out if you are ready for the battle. This, he says, is bringing out the best in real leaders. He lives for the moment.