The Cost of Building the Right Thing | AI, Speed & Discernment at ServiceNow

Welcome, everyone, to another episode of ServiceNow Insights. I'm your host, Bobby Brill. Today, we're getting into something that is the most important conversation happening inside any company that builds software today. AI is making engineering teams dramatically more productive, 10 times, even 100 times.

And that changes everything because the systems around engineering, the ones that make sure what we build is actually good are racing to share the lead. And the cost of building the wrong thing, well, that's also climbing exponentially. Today, we're going to talk with three people here at ServiceNow who are tasked with winning this race, and they have three very distinct points of view. Gentlemen, before we get into it, please introduce yourselves.

Hey, Bobby. Thanks for having me today. David Hoare, Group Vice President of Digital Content and Design. I've been at ServiceNow for five years.

Thanks for having me, Bobby. Anand Tharanathan. I am the group vice president heading up product research and insights to service now. I've been here since 2022 August.

My name is Stantley Davis. I'm SVP of design. Next month will be one year for me. So, David, you've been thinking harder about the productivity math on AI than just about anyone we've spoken to.

But before we get into the tactics, what's your philosophy on AI? Talk about the lens you see all of this through. like from the moment that um i think it was like chat gpt2 burst onto the scene i was like this is something different right i've been through the the internet boom saw like crypto all these things but this is something that i saw and i was like this is really groundbreaking and game-changing and things have only accelerated since then i think it unlocks huge amounts of of potential and one of the areas that I come at is from productivity.

So I think that there is so much that we can do and so much that we're unable to do today. And with these new AI tools, it's given us really a way to start to do stuff that was never before possible. When I say never before possible, probably economically viable, right, is the way that it is able to reduce costs for us in many different ways opens up avenues for new things that we would never be able to do before. So I think about one of the big teams that I run is creating product documentation.

And having humans understand what the product does and write all of that is a very, very expensive proposition. And so there's only limits. There's limits to what we can do. Whereas with AI, not only can we do all of the documentation, but we can make it work for you.

We can write it differently for Anand. We can make a version for Dantley. So those kinds of things, I think, is just one example of where it's opening up new doors for what's possible. Now, Anand, you're the person in the room who consistently asks, what does all of this mean for the human on the other end?

What does that look like from your side of this? So I have a perspective. So obviously, I completely resonate with David. I think as a technology, what we're seeing is pretty unparalleled.

It's very exciting, nerve-wracking, and sight-inducing. How are we going to call it? We're already starting to see one person, two-member, three-member startups that scale from zero to 100 million plus ARR and scale with just three people. So you see that kind of stuff happening around you.

It's just awesome to watch. And Dantley, you're a designer. You're the creative voice in this conversation. When you think about AI and creativity, where does it take you?

As it relates to creativity, I'll take you back to when I was a teenager and I was using 3D Studio Max, which is a 3D creation software, pretty heavyweight software that was used for a video game industry and movies. And I, as a teenager, spent hundreds of hours learning that interface to try and create something pretty basic just to navigate around the system. And you fast forward 30 years and use MidJourney and you don't have to know an interface anymore. You just prompt it, what you imagine, and this amazing thing just becomes rendered and you shortcut the need to try and wield a computer interface.

And there's a saying that art imitates life and vice versa, and you mentioned science fiction, and it reminds me of the Marvel Universe, the Avengers, Iron Man, and Tony Stark's relationship with Jarvis, where when Tony Stark is in the Iron Man suit, Jarvis is the connection point of all the things necessary to make Iron Man a superhero. And AI, to me, offers that possibility of human superpowers and lowering the barrier to entry of having to learn and wrestle with interfaces, drop-down menus, buttons, all the CTAs, and just get to exactly what your creative mind is thinking without a whole lot of manipulation on the computer side.

With art, it's meant to elicit some type of emotional reaction. And for software products it is meant to solve a problem And I imagine and I seeing this today in our day work that as we become faster with our productivity suite leveraging AI we spending less time manipulating the interface and more time obsessing the customer problem. And I suspect that most of us want to be in that zone because you have more option value to create many more solutions than just the singular solution.

It gives you more time and space to iterate and learn and get feedback from your customers. And ultimately, that's what we're collectively working towards is to solve those customer problems. Anand, the customer has always been the North Star. How has that changed?

Or has it changed? What AI has brought into this whole ecosystem, if you want to call it that, is it's really helped us accelerate the means to the end. So the whole process of building product, what used to take weeks, months, you could do that, you know, in days, right? So that's super exciting.

With that said, the whole concept of building products for users and customers hasn't changed. You're building a product to find a product market fit. You're building a product to solve a problem for a customer and end user. That aspect hasn't changed.

And David, engineering teams are building at 10 times the speed. are even 100 times as they were two years ago. And they're building super fast. Now, you describe this as a survival problem for everyone around engineering.

Walk us through that. So one of the things that I realized a couple of years ago is that one of the first areas that those big productivity gains are going to come is with engineers. So engineering is, I think, the leading industry in terms of AI adoption. And with that comes huge productivity gains.

They can build anything. they can build way more, like 10 times, 100 times more than they could before. But just because they can build all that stuff doesn't mean it's going to be good. But where I was coming from is if they're building all of this new stuff, what are all the systems that are around that and all of the roles that you need around the building itself to make sure you're building the right thing and make sure that our customers can adopt it?

So if I start with documentation, right? It's the number one scaled resource for customers and partners to be able to use service now. And if our engineers are twice as productive or three times more productive, they could be putting out two or three times more product that needs documenting, right? And if we're already struggling to keep up with where we were, think how much more we're going to get in the hole without finding ways to scale that work.

And so I think there's kind of like a survival that we need the AI for is like the product experiences are going to get worse unless we can use AI in those supporting functions around. Another example, product leader Andrew Eng. I think he was chief scientist at Baidu, founder of Google Brain. He said recently he's seen ratios of engineers to PMs changing from 1 PM to 14 engineers, to 1 to 7, to 1 to 1, to often he's got 2 PMs to every one engineer.

The cost has dropped that much, but the cost of building has dropped that much, but the cost of building the wrong thing has climbed exponentially. And I think that applies to all of us. With the user research side, understanding the customer and what their needs are, design, making sure we are building the right thing and it works most effectively, that discernment is where I think we need more productivity to keep up with the AI productivity. And Danley, how does that dovetail into the creative output?

One of the mantras I followed through my career as a designer was design the thing that was asked of me and then design the thing I thought was better. And better meant that I had enough time in the day to actually do the work. And oftentimes that meant very long hours and tinkering well into the night. That I understood the customer problem well.

and I was only able to output around those constraints. And so in any given project, I might have a clear hypothesis and a number of concepts around that hypothesis. But I am looking around the corner and I might see another customer problem. So that might orient me around a couple of different hypotheses.

And then I have a few more concepts to try and prove out product viability. But that takes time. These days with AI, rather than taking a week to create nine concepts for exploration and trying to find market viability, designers can create those nine concepts in an hour. And so it just speeds up the ideation process that was limited by bandwidth in the past.

and it's not to say that those nine concepts or ten concepts are great but to get to the good idea I believe that you have to have a lot of bad ideas and you go through the discernment of seeing something and saying okay that doesn't work really well because of these factors but there's a there's a kernel of amazing in that one concept that I might want to pull into to another concept. And AI gives us that ability. We still have to slow things down a bit. Like once you have all these options in front of you to pause and edit and curate Some people call that taste Some people call that judgment And I think as an industry we starting to realize in this first chapter of AI software development that slowing down becomes part of the process.

And through the slowing down of talking to your coworkers, talking to customers, you're able to get the input and iterate. And then that allows you to speed up in the next cycle of software development. Whereas maybe today and six months ago, we were just running, but not clear where we're running to. Everything you said about what we should do, I like 100% agree with that.

Can we do those same things faster than we could yesterday with AI? Can we iterate with the customer real time? Can we provide the engineering team with daily demos and daily prototypes, like five different versions of it. And we can, right?

We saw demos from the team today. So I think we're all aligned on the work that needs to happen. The whole notion of slowing down is definitely relative. I'm not thinking of months of indecision, a navel gazing.

Maybe it's 15 minutes to just align. But, you know, it reminds me of the evolution of music from going analog to digital and the introduction of beat machines in the 80s and the rise of hip hop, where you're able to create beats and rhythms a lot faster. But there was a tastemaker that was listening, maybe taking a beat, no pun intended, to find that right hook. Yeah.

And even though all the mechanisms around that activity went faster, someone was taking the time to choose the right sample. Yeah. And that's kind of what I mean by slow down, that we come together to find the aspects of what we're building that we think are great, and then we go. It'd be like, that's the one.

Yes. Like the discernment. I 100% agree with that. You know, discernment is a word that keeps coming up in this AI conversation.

But discernment at scale, that's a different problem. ServiceNow has years of content and products and documentation and a lot of great assets. We've made a lot of really good stuff. Now, David, I want to ask you, how does AI help unlock some of those opportunities to use all that we've got?

And the other side of that question is, does it freak you out to use what we've got? One of the initiatives I lead is like content governance. So we have a bunch of different teams across ServiceNow from marketing to the product team to learning. We're all creating a lot of content.

A lot of it is focused around how can we help our customers be productive and successful with ServiceNow. Now that content atrophies. It gets stale and it gets less useful over time. As our products move on, if the content is not moving on at the same rate, it becomes less useful.

right and where we are now is like an AI consumer of that content is just as important as a human consumer of it and where governance is important is if you're not managing that content life cycle and aging out the old stuff you're going to end up with hallucinations and drift and all those kind of things the good news is we can use AI for governance as well I mean it does sound a little bit like the fox guarding the hen house. But as we talked about today, like we had an engineering lead come and talk to us and she was talking about the proliferation of artifacts.

Like there is just so much stuff and how much time you need to take away from building to reviewing. Like you constantly need to set aside time to review everything that's being created. With human governance and things like that, you're never going to be able to do that because content is just exploding. So I do think we need to be using AI and then have, you know, rigorous structures in place to manage the content lifecycle of these assets so that we can get rid of them out of the systems to make sure that the AI continues to perform well.

Maybe it feeds into quality. It's like the AI is only going to be as good as the signals it receives. So like you've got to feed it. If it's in databases, tables, or unstructured content, you've got to make sure that's good.

Otherwise, you're going to lose user trust because your quality declines. David just said something about if you lose signal quality, you lose user trust. How do you think about trust from the human side? Trust starts with ensuring that we build a product that resonates.

There has to be an aha moment. The first smile experience has to be awesome. whenever the user needs the product has to work and it has to be performant. All those things create trust.

Now, when you think about human AI or human automation models, there's always a concept of how do you build the AI to circumvent the limitations of the human? And when you create or leverage or capitalize AI systems to circumvent those limitations and build experiences that can bring the human AI system to a much better spot, that in the long run builds trust. So governance, yes, but I think there are bigger aspects of how you build a product within the AI human system that are bigger predictors of trust.

And Dantley, where have you seen AI surprise you or do something that you didn't quite expect, something you didn't see coming? The example I'll give in the context of good is that we have moved to a new UI tech stack leveraging some new frameworks and using AI to develop the code for those tech stack leveraging some new frameworks and using AI to develop the code for those tech stacks have meant that we have more or less feature completeness in terms of how those UI components will work.

Where in the olden days, humans had to define all of the capabilities and permutations of those UI components, and oftentimes it didn't meet full desirability maybe of certain business use cases. So that work that's happened with our new products has happened incredibly fast, and that's been really amazing to see. What concerns me without governance and one of the benefits of governance is some degree of predictability. And without governance being within our, say, design systems, as an example, if teams are using AI to build a bunch of UI components, the issue of feature incompleteness can happen much faster than when it was just humans creating components that lacked those features.

And we're having to manage that now because we have a new AI powered design system. We have business units that are coming online to develop new products using those components. Those business units have very bespoke customer needs. And without the systems and the teams talking to one another, and then with the power of vibe coding and AI software development, now you have this huge rush of new artifacts and tech debt that you just push down the road.

And there's this hidden cost of tech debt that happens where initially the romanticized view of AI happens because you're seeing all these things happen a lot quickly, a lot faster than it did in the past. But then you move downstream in a couple of releases and now you have all this hidden tech debt. So governance potentially can help us address some of those concerns and ServiceNow as, and I assume every company that's building software right now is trying to figure this out.

And to David's point, if you have exponential growth in artifacts, but you still have the same number of people reviewing it, now you have a bottleneck. So we're all working through how do we get past that? So last question for all of you. And it's the only forward-looking one.

18 months from now, I know that's the standard question, 18 months from now, when we get this right and better, the discernment, the governance, the human element. What does that actually look like? Anand? The biggest thing for me is that we need to ensure that the human factor is not lost in translation.

The human element is and should always be the central part of any man-machine system or human-machine system. That is my wish and hope that we do not lose sight in this mad rush, if I may call it that, important one, that the human always, always, always essential. Yeah. Now, Dantley, AI has to be all that for you at this point.

I mean, your fingers don't even have to do the work anymore. That's right. And it speaks to what I hope we can see collectively in the next 18 months, which is that as I've spoken to customers, and I think the listeners of this podcast could probably attest to this, we're all struggling with leveraging multiple complex software systems to get our work done. Whether it be unstructured or structured or licensed or personal, there's just a lot of very challenging and difficult navigating of systems to get work done today.

And I believe that the promise of AI and the integration in our workflows will enable any person to get their work done without getting frustrated with the interface, without pounding their fists on the table, without cussing at their computer. It's just the thing that they want to accomplish is as easy as speaking it, as if you had a very intelligent, wise colleague next to you that can will this to happen. And I suspect that will make the work lives of millions of people a lot better.

Now, David, I know your answer is going to be about where the rubber meets the road. So what's your closing thought? I think it's pretty easy. Customer value, right?

My team exists so that our customers can be productive and successful with ServiceNow's products. So over the next 18 months, I want to see an increase in customer value that people are deriving. Because the more value they see, the more of our product we're going to buy. The more ServiceNow share price goes up, the happier I'll be.

So there you have it. Customer value. And isn't that always the answer? And it runs to everything that we've talked about today.

The philosophy, the speed, the discernment, and of course, the governance. I want to give a thank you to our guests, David Hoare, Group Vice President of Digital Content and Design, and Anand Theronathan, Group Vice President of Product Research and Insights, and to Dantley Davis, VP of Design. For more information on what we talked about today and for past episodes, check out the show notes and hit subscribe so you never miss an episode of ServiceNow Insights. I'm Bobby Brill.

Thanks for listening.