#141 AI Pat Works Here Now: Why Agents Must Follow Human Rules with Pat Casey // CTO @ ServiceNow

Hello friends, this is the Alphalus Podcast. I am your host Tobi! The goal of the Alphalis podcast is to empower CTOs with the info and insight they need to make best decisions for their company. We do that by hosting top thought leaders and picking their brains for insights into technical leadership and tech trends.

If you believe in power of accumulated knowledge to accelerate growth Make sure to subscribe Plus, if you're an experienced CTO. You will laugh. the discussion happening in our Slack space where over six hundred CTOs are sharing insights or visit one of all events. just go to alphalist.

com to apply. Welcome to The Alphalists podcast. I am your host Tobi and my guest today runs engineering at One Of The Biggest Enterprise Software Companies on the planet. Something like eighty five percent.

a fortune-five hundred are your customers Thirteen billion in revenue and he is the CDL of a company called service now. And, He has basically there from the very beginning which it's rare like. I think if you see publicly listed companies then Very often The founder gets replaced by some more like let say enterprise and stock ready CTO. Welcome Pat Casey.

Thank You to pleasure be here So maybe we start very early Even before your career. let's say, Pat what is your nerd path? What was the earliest moment you thought computers are kind of magic. Oh it's a good question!

I actually got into it. so i'm naturally an incredibly strong introvert and can fake well at this point in my career but like have been all my life. And one of the awesome things about computers is you can play with them by yourself or do something By Yourself. so I actually got Is an Atari?

four hundred Which i think, i got remember that right had a little checklist keyboard and it did like storage. was You plugged A tape player into It. And it Was such a low band width record for storing bits there but you could listen to it you Could hear it sound Like Burge chirping. It's all analog, and it really couldn't do much.

But it was a computer! And I taught myself...I think that I had Atari basic at the time and it wasn't super useful but got me interested. Then I went to PC Junior and used almost exclusively to play computer games.

I was a huge player of very, very early computer game called Wizardry which i'm going to date myself. The name continues but the Proving Ground is the Mad Overlord that with me at age twelve or thirteen banging away But in terms like how felt using it for something functional. I am relatively smart guy In school would get channeled into you know, the smart guy things. But I never liked called abstract math as an engineer tract.

I wanted to be an engineer and always felt myself like why am i solving this massive foyer transform? Like Why Am I Triple Integrating This?! I can just model it! So for me that's actually...

To Me It Was Easier To Algorithmically Model Something And Just Solve It That Way Than It Was To Do The FFT Or Whatever. and that got me conceptually into it. But my first real job, someone paid me to do something with computers... I'm a failed graduate student but during grad school like i just wanted a job too.

so I was working for company called Aldis who made like page maker back in the day And they got bought by Adobe Jockey. or see me like a floppy disk jockey. I would go to your desk and i'd install software on floppy disks. But remember going to my boss, I said you know what?

Like we got this huge company now We just merged with adobe! We don't even know who works here...like..we need system to keep track of all the employees.

And she's like yeah it seems really useful thing....i think that can write for us. So i fired up microsoft access and had no idea what was doing but built The first, I don't know if it was the first. It's the first I knew of employee tracking system at Adobe and then its like.

oh we can do more with this! We can track bills of stale with a lading dock And we can track when things come in and put little labels on all computers to keep track. So i just got into basically because the job had needs that felt as though they could probably Not that dissimilar from a lot of, I guess you call it fairly bright but young people when your young. You don't realize the current state.

so just like i can do this myself? Yeah and if somebody's right in this case i was right. So basically introduce ITAL at adobe on very early stage. But yeah inventory management tracking stuff we take that for granted now because But it wasn't, you know in the middle nineties like keepin' track of stuff.

You would lease computers from your Dell or somebody and you'd leased a thousand. I mean at least we had to give them back and be missing two hundred. It was not that unexpected. And then he just paid the penalty clause but this state-of-the art in some very high functioning companies At time is an unsolved problem.

Then basically with Service Cloud from the start, was it your friend? The founder of Fred Lottie or how did that happen. I actually knew Fred in between my graduate school days and my service nowadays... I worked at a never-company called Peregrine Systems and Fred was the CTO there..

I lasted maybe two years then I left but Fred stayed little bit longer But he lived in this same town i lived. So i remember I was actually at a convertible with the time and driving down the street. somebody walked in front of me. I guess there are crosswalk but anyway, i thought they were abrupt.

so i honked at them. when i flip off like wait a minute that's fred levy. i just flipped off red levy. So my friend pat you know we're stopping traffic or chatting he is like i just started to company should come by see what i'm doing.

I took thirty days eventually went buy And he had a little office, I think it was like upstairs from the restaurant because one of his buddies owned the restaurant. He showed me at that time as super-super early version service now called Glide and i sort didn't get it Because At The Time Like This Davey Art Enterprise Software. There Was This Push Toward. I'll Call It Sort Of Bite Perfect Full Fat Configuration.

Like you would have a tool with like, a forms designer and look at... You could do anything with it. You can put the button on the ceiling if you really wanted to. And Fred's insight was way in a couple but one of them is that none of this mattered.

What we want to do make it easy for the common stuff. The early version of Glided did made it easier to form lists and navigation which are just the common stuffs in enterprise software. And the other key thing of course Fred had was Internet Deployable. It was all browser-based.

And it's a SaaS product back when SaaS is new, but I liked Fred and I sort of wasn't living my best life sitting at home. so I'm like. well you know fred yeah i'd be happy to work with you on this...I love the opportunity.

um and say I was the first person other than fred uh to work on the codebase..and that was I think early two thousand five? I'd have to double check my dates. but north of twenty years ago.

so crazy! Super strong introvert as you describe yourself leading a company with how many engineers? like ten thousand or. When we care for me bill mcgermain release the companies to see you and cto is different roles.

but actually up until about by month ago yeah i ran all of engineering. so which was about, It's about ten thousand human beings, about seven thousand of whom wrote code or plausibly wrote code. So if I looked at like how many licenses have like windsurfed today by a lot of us and it's obviously very different um... And i won't speak for all introverts out the world but i'll speak for myself.

You put me in a crowd of peers In some weird environment? I don't know. Like I feel very uncomfortable Make sense. I don't know these people, i dont.

where am? it just feels alien. give me a role in that context and i feel totally at home. um...

i can be the photographer of wedding. you know. i could be working to sidelines at an afl game In work contexts like when your cto have a role everywhere u are. So oddly enough for my personality its psychologically easier uh..

To be a CTO Then it is to be one of the people just bracket file for lack of a better word. Just because in any environment I'm at, and there as the CTO if i'm meeting customers not they're just talk to the customer's meeting employees and Talked with the employees. hopefully keep morale up listen their concerns. It you got a role?

And from me that super important and uh...I guess that roll also changed like ten to fifteen times dramatically throughout your time at ServiceNow? Is that like roughly... Yeah, I mean yes and no.

So on a high level yeah when the company is really small you just everybody. there are two job titles. There's engineer in this Fred Lutty. i mean those were my two jobs titles And it was not fred Lutti.

um You can kind of make it work about twenty Twenty engineers, we started having to like divide the code base up a little bit. It's like alright We can't all work on everything because like the commits collide all the time and were breaking each other's code. And so there that was a bit of a change in role for them. I'm doing Everything too okay?

Like i have A LITTLE BIT OF A PLAN LIKE I'M WORKING ON THE DATABASE RIGHT NOW. Before That WE ACTUALLY HAD AN INCREDIBLY SUFFICIATED SYSTEM. THERE WAS A STUFFED FISH. SOMEONE HAD GOTTEN AT THE CARNIVAL AND IF YOU WERE WORKIN ON SOME OF THE the guts of a code base, you put fish on your monitor and people know it's like Jay is working.

It worked. at some point I think we thought were going to get different totem animals for different parts but we didn't. How often was that fish in your monitor? Pretty frequently I don't know.

Like, i can write ui but like it's not my passion. so i tended to work more on the back end code um and then the database layer um processing...like..I was a guy who would figure out the profiler in.

you know? Like, we have our own date parser. You know, the jvm has very good robust date parsers dates that are system only three possible formats and the JVM's date. parser threw off a lot of garbage, which mattered to us.

So like I wrote in Java but I wrote our own day parser was produced less Java. Like I was that low level guy not exclusively. But if there is something needed get done. it would usually be me.

Suppose other inflection point. you got about one hundred people broken down into teams But the ability to just kind of know what everyone's working on and see who has got the fish breaks down. And then you start needing some process in place, and perversely my experience is that it gets less efficient. so like...

You get this sort of peak productivity at a hundred engineers are we're gonna put some processes here? It slows you down! And if your getting less efficient probably until you have maybe one hundred twenty or one hundred thirty engineers Then come up with the other side of a trough Volume of engineers at the problem which offsets. The overhead And then it's some level that you can ramp that for a while.

There's various changes we had to make to our release process along, the way. but the underlying structure Of like hey You got teams of engineers. they're more or less seven plus or minus They work on a backlog and check in. That paradigm has largely been unchanged Fifteen years, although it's definitely starting to change with the rise of AI coding.

The whole paradigm is breaking down or evolving. a different word Yeah evolving well changing dramatically I'd say from my perspective but It really depends on the company at the problem. But maybe let me step back. so you already told us that You use Java, so I guess you do have a Java monolith.

or is it all like microservices now? Like did that dramatically change or. So the core service no product. and then one that Fred Luddy wrote all those years ago in this sort of the core thing that we've been selling It's at some level to metadata processing engine like it doesn't Do anything until you give it a ball of metadata.

And the metadata defines everything from like with screens look like to navigation paradigm to work flows too. Like everybody's password and user ID, it's all. meta data processing engine is all written in Java just like you'd expect. the metadata itself usually stored in a database in rows and columns but there are scripted metadata And the scripted metadata, like a workflow.

you might have a business rule. It's server-side JavaScript and we ran it in... Still do runnit In somewhat hacked version of The Apache Rhino Engine. I say somewhat hacked because We changed some behaviors Of it and tightened up Some of this security.

But if You look today i mean nobody writes apps Like that. its very I don't know two thousand four. So if you look at the newer stuff we're doing, it's all k-h based. In a lot of cases its daughtered to the main Java code.

At some point the java code wants to call other service and it just goes over into KH cluster And there is an effort where rolling to get more and more out of the Java ball And into the world of case. I can iterate it faster. Um, i can scale it better in some cases It's a very I mean this sound like an engineer but I am. its uh is A Very cross linked code base with a lot Of subtle implied dependencies.

So it's a much harder job to tear it apart than you might think Like on paper like anybody who? Anybody of any senior is like, I know what to do. We'll break this thing apart. It's like all right.

So it was that easy? Like none of the world's big monolithic code bases would still exist. like a lot of cases Its its process But we're making good progress and frankly anything new we've written for the last probably four years. It's all.

it's all Kubernetes And its language is there very a lot AI stuff with Python A lot of bits node. We're big uses of Ruby. I don't know why, but we just use it a lot. So far as i know ,we don't have any Rust lurking anywhere even though It's the cool language right now But not yet.

let's see Right and not over here. and architecture wise like you described Kubernetes And yeah some database sitting there where will also dig a bit late. Dig down a bit later What? You guys made this Contrarian bet.

Let's say that for a very long time at least, you were like mostly single tenant and every customer got their own dedicated instance. Is that still the case or did they change? It's still mostly correct. so if you look at as service now instance it is cluster of JVM talking to database with low balancer in front.

It's not a one-to-one relationship between those software entities and hardware. I share the willies out of underlying hardware or virtual machines depending on where we're running you, but if your AT&T pick a big customer... You can go and point to the JVMs in the cluster which are servicing those requests, then there is a bunch security so that they only talk about database etc.. From an operational standpoint this it's actually some of the parts that are wonderful because long as you've got to database.

You've got the customer. I go one thing to back up about one. think a fellow over. i've gone on thing to deploy like.

that is the universe, um its also scaling pinch point because like matter how big your databases at some point you cap the ability for building do work. so thats the core architecture. If you look at some of the newer services we're doing, like AI Services that's a share tenancy model. So I've got to cluster depending on where your running.

Sometimes i use Frontier models and sometimes internal models. but if using internal models... I gotta cluster servers with Running Triton cooked up to GPUs And then I have a proxy service in front of it. You make a call over there and route your prompt into the right model.

It returns But you never have to, it's not a one-to-one relationship. Similarly for things like the UI refactor I mentioned we were doing to get that out of the monolith because there is no nothing that needs to be persisted right? Like in most cases? yeah um There will be.

It has been this way. i mean i've got enterprise customers who've got something between contractual expectations and just like expectations That if they're running an particular environment thats where their data it doesn't leak out somewhere else. So the places where we have a good opportunity to change the tenancy model are usually when you stay in the same data center and its not persisting any data, When I want to bust up of data at rest... A lot times they need people sign new piece paper which can happen.

but that's one case is enterprise has constraint. probably isn't there consumer space You know, do what you think is right. Hopefully What Do You Think Is Right? And that like leads to a crazy number of databases Like.

I heard That it's like roughly ninety K Databases like ninety thousand database or eighty five thousand something like that. It sounds about right and its no human being ever looks at them. They're all managed by automation. And yeah, I mean every now and then something goes wrong.

the human being probably like swoops in to try to figure out what went wrong. Yeah It's pretty rare and In twenty five billion queries an hour so we would have why if We probably gave you that around. she expected it be a little higher than that. I Mean the way service now works Like any request do you do?

I mentioned its a metadata processing engine like The metadata itself is in the database. So if they wants to know How did like you make a request to order new laptop and it's got work flow around what laptop do u get? Are you authorized to do that, and what vendor I buy from. That is all metadata in service now And we try to cache but caches don't always hit.

So if i get a cache miss on stuff It goes into the database To leave the meta data Even though U as user may have thought Hey! I saw picture of a laptop with button said order Behind the scenes. we may issue four hundred queries to try to figure out how to do that for you on your behalf. So there's a big amplification between like the number of screens.

We render and the number. database operations, we do And most of them are incredibly highly optimized selects by primary key. But it's a lot. I mean numbers go up Up into the right.

It's also why? We bought a database company. So we used to be my SQL company with big NODB, we transitioned to MariaDB because of the relationship and then ultimately there were challenges scaling it for our workload." And you said on board of MariahDB as far I saw right?

You worked with Monty... It was Monty that actually blanking on CEO at this time is not Monty. but Monty was CTO there and he knows that code base better than anybody. Yeah, I wrote a lot of it!

But his heart-of-heart was in like hey... He wants to have everyone the world succeed and love mySQL codebase MariaDB codebase. And MariaDB is the corporate entity was trying to build itself into a more full featured, hey we're a data repository company. We've got a variety of engines and we've got bunch tools.

you come with us. your data problems will give the solution. so they want it be like a MongoDB or something but obviously SQL based. whereas Monty I think in his heart he's that interesting maybe great for the company.

i wanna work on my SQL code because this is where love And I think these days what he's working on there, it's more supporting the traditional MySQL in a DB code base. There and I talked to him maybe nine months ago He seems actually pretty happy with where the role is right now. Okay cool! Then you built your own system or acquired company.

that was like the inception of RaptorDB which is your database? Yeah Yes, so we bought a company called swarm sixty four. I was actually German Company and what they had built? Was a column store on top of Postgres.

But with the problem They had solved which i thought it was A hard Problem was that their Column Store was always in sync With The underlying be tree based Database And So They Would Transparenly In The Query Optimizer Be like oh That query can be efficiently handled via the column or indexes. I'm just going to delegate there, whereas this other query in a columner cancel that we're gonna go into regular beat. tree index is but it was all transparent like you'd just below the sequel layer.

so he wasn't application developer just fired SQL at it and your dislike man. like these reports got a lot faster. This Is Amazing! But...

Fifteen person German team and nothing with the Germans, but like fifteen human beings can't usually take a database all the way to enterprise scale And it was optimized for whatever they thought. The workload was so when we bought them were like hey We love you guys working. give me some more resources. But your workflow target is now this service not work load.

I want you to make as best possible database for the workloads that shape like our workload. It's my knowledge. like every one of those core engineers is still with us on their excellent team, but the team got bigger and The workload shifted. But about two years later we started rolling out to our customers under the brand name of Raptor DB And that was actually named by Bill McDermott Which he didn't.

actually if you look there's a picture of an eagle We use it as a raptor But to this day, I'm not sure when Bill named it. He was thinking of that kind of raptor or like the velociraptors from one of those movies where the dinosaurs kill everyone but we went with The Eagle which felt better for marketing than a velocirapter. So database became like a marketing thing? Was that because your system was too slow or?

And is it ultimately then something that you upsell to customers, Or like how does it work. Like so there's two flavors of Raptor. There's raptor standard raptor pro. they're both quite a bit faster than maria db was Largely because they use modern high core count computers better and there are just some.

InnoDB was written back when most computers had one core. So if you look at its core memory structures and architectures. It just doesn't work well. when you get, I don't know maybe twenty quarters north of that it just spends all its time like spin locking um...

And it may be some smart guy like Monty is gonna figure out how to work around that but in the time i was on the board we couldn't solve it. But very few workloads need that. however ServiceNow has one those workloads which needs. So what Raptor Standard does is it gets you, based on Postgres that architecture just handles high core counts better.

And it doesn't save you some storage but that doesn't matter to the customers and then everybody gets it. I think on like ninety seven percent of the way through converting my ninety thousand databases over to Raptor standard It's got some small number if MySQL still in there, MariaDBs Raptor Pro, you do pay for. That adds the column store and that adds a few other features which are relevant. if you're like really big and care about your reporting performance And I think Bill has him because he did something similar to SAP right?

You could pay for HANA. Yeah. So I think in his mind that was of course you should pay. let's say it's IP playbook yet yeah.

But like i said everybody gets standard. their standard is you don't pay for it. You just get upgraded and run better." Okay, okay.

And did she ever... I mean your in a way running. then Postgres? Like how big is the difference between like your own proprietary form set or your own proprietory technology and core Postgres ?

Is that huge ,or at feature level than where Something different or made some level word downstream fork of Postgres so we can still pull the upstream patches. There's a lot of additive code because we've got the whole column store Which is it's all our code. and then like I could get like The database architect to walk me through if there are parts of the postgres code base Like the optimizer where we had to come in pretty aggressively and make changes But we generally were pretty good at keeping keeping up with the down stream polls And we do contribute some stuff upstream, too.

So if you look I mean We're not like a top five contributor to Postgres but we do and the other place actually where or you might surprise You were decent-sized contributors open JDK. Because we run Open JDK internally and we optimize it and fix It and tweak it. and you know i've got A team of very smart engineers who worry about Stuff Like? The use of perm gen Or shared space all that fun stuff.

Okay, yeah cool and did you ever think about also releasing your fork as open source or? Releasing the Raptor add-on? basically I actually don't know like it's been tossed around internally. i don't no anyone got too serious but we're pushing some of the code upstream.

so its not like were hiding. The stuff that we might be concerned about. open source, frankly is the column store because we do feel like that's RIP. But let the OpenJDK stuff has literally opened JDK so push everything upstream.

and usually these days your committers take our stuff but every now in them will get an usual open-source. hey it's a good patch. no its not! It just...

the way the world works. Can you imagine? Then slowly coming to AI topic which I think Yeah, keeps us busy and productive. And hyped these days right?

Like everyone has a different opinion. that's like kind of always interesting to hear... To sync up on those! So I think in an interview that i saw in the year twenty-twenty four you basically made big investment into windsurf licensees for all your engineers.

um.. You said this brought ten percent productivity Back in the days, not absolutely game-changing from where you were before co-pilot. How did that evolve? From there like how hyped are you yourself?

Like do your run any Claude on your own computer right now? or...like..how does it look for you yourself?

and yeah, how did that evolved? so I feel we're sort of a third generation of coding tools. You know, first generation was called like the early GitHub co-pilot. It's like slash.

slash give me a function that tells my numbers prime and it'll spit out their function for you. And I think they were little labor saving. but it was just better version of auto commit or autocomplete Second Generation. Yeah we're big investors in Windsurf.

We did not buy any stock But we bought another product. I bought about seven thousand licenses. We trained seven thousand engineers on how to use it and we could demonstrate, we got about fifteen percent of a productivity bump out of that just measured as stories per engineer for sprint. you know very little by sprint but around numbers fifteen.

And that's consistent frankly with what i see when i talk to other at scale companies. so they're big tech companies Big consumer products companies who've got a lot of engineers. What you see though is it's not that everybody got fifteen percent faster, yeah a lot of people frankly didn't change very much and then you've gotta small subset of people who really just clicked with them. And their code productivity goes up in five acts or six act.

And there's also another tool we added to the toolbox. Um, that nine months ago we also added a quad code to people's toolbox and said hey look like if windsurf makes sense use windsurf got clawed code. And what? We see is for the people who are still doing sort of I'll call it code centric IDE based development Like where you're looking at the code.

And your thinking in the code It's gonna be Windsurf ninety percent of time. But if you shift to sort of that new hate these word vibe coding because its so overused. but i'll Call it output centric It model where you're prompting and then looking at the output, and then prompting again. that is usually done right now with quad code inside of service.

Now although we had were actually looking at other options just because I like anthropic it's a good tool but like We should look at other stuff too And it's usually CLI based And there's a super strong Pareto chart on engineers in that some of them seem to click with the coding paradigm and you get this. Some people see ten acts, I haven't necessarily seen ten X but you can say five X lines a code written. maybe someone is it ten. I've pulled the data?

But they click and there just off to races And a lot of people just don't click. um point of view in the industry, which says like it's just resistance. Like people everybody could click if they just got past their fear of changing paradigms. I'm less sure that to be honest because at least for me It's a different way of thinking.

and The best analogy i have is an imperfect analogy Is if you're good at chess usually learn to play person-to-person. You've got a whole one chess board in your head, figure out what's your opponents playing the person? But like there is subset of chess players who are really good at playing five boards at once and it not always that people who were bested across the board probably does overlap but its'nt always. AI coding If u get to that next level is like playing five Boards of Chess because you have multiple prompts spending time They're interrupting you when they finish, and then you've got to react.

To them. it's very different from the traditional coding paradigm where your in-the-code I mean Traditional Coding is almost like an Autist Paradise because its incredibly focused. You are holding a whole code base on your head while even if that tiny little bit That why the traditional to non engineer? You don't ever understand why an engineer gets so upset when you interrupt him.

he got a quick one for ya. do you want to cup of coffee? i just blew up in hours worth my work is your lot. the whole lattice of logic collapsed and your head.

So I do think thank fear expect, i don't think everyone who was good at the old paradigm will be good if new paradigm. And i think there will be some people were just not successful the old paragraph or hated it or he didn't click with them, who do click? With the new paradigm. So I think it's going to reshuffle the DAC of Who is a top productive product person and It's not always gonna be me.

Um...it could be someone else. so yeah, it's uh potentially more the More product-driven people right like. um if you have to turn from Like Someone who receives instructions Or requests To someone who Is actually formulating Them And its like A natural Pivot to becoming more like product driven potentially right and really thinking about a customer problem.

I think that's not Like what every engineer loved back in the days, right? I Think that's true. i think we have still found. Because i'm like i said i got seven thousand engineers And a lot of them are transitioning To these new whip.

The most efficient model is if you get an engineer who clicks with the new coding paradigm It has good product taste. they're just off-to-the-races. A lot of times you get someone who's a good engineer and click the new product coding model but does not have good product taste. in that case i've got to kind of marry up with designer or product manager when your working like little group.

But they all model. where you'd have like engineering was the scarce resource any like seven engineers to keep. it is either busy are seven years keeping project manager busy, And like I said, in a lot of cases if you've got the right engineer who's got that product taste. You don't need The other two.

or frankly If you've gotten really good product manager Who clicks with the vibe coding? Like what do you need an engineer now like just Claude code and knock your socks off? Perverse example about all give which is imperfect but it personal for me as my daughter. Very smart lady Mechanical engineer Does not like coding Irritates her loves products though.

Um, and she that's why she probably mechanical engineering is opposed to double e. but She had some summer project last summer where? She was supposed to build the ui for some trading software with some bank That it was interning at. And she called me up and she's like are you complaining?

I'm like. what should i do? well You kind of taking their money. you got a.

give them what they're paying before he's like I get that. But um What Should I Do? I'm Like. Well Here's Codex, here is Claude code.

Like heres some YouTube videos on like how to do it. You're a smart lady you'll figure out. and like week later she says this the best thing since sliced bread. All I tell when i want builds in.

It took annoying part of process away from her but left her with. that was validating for hers. She always liked building products. So think there will be.

I think we're shifting as to who's gonna be successful and it's not always going to the same cast of characters. We had five years ago, back tour your company an agentic. what you take on agents? like agents not in terms off like engineering agents like cloud code but agents for the rest of us.

I mean, basically you're dealing with IT service management and many large organizations. So is there like how many agent real agentec use cases are there for you? And in the real world out there How much of this true? what other companies telling us about it still happening down the road?

So what are the coding agents? I feel like we're that's generations of how people are applying AI in about the enterprise software space. The first generation, I would say it's uninspiring but a little uninspired. Like i'm going to summarize an email for you or like your working at case and Bob just handed off too much quickly summarized all work Bob did so he want to scan through his history.

I save you five minutes there. ten minutes it's real is real savings and make a more productive but not going to change your life. And then you saw maybe eighteen months ago, there was a push towards like agent frame works an it was the little field of dreamy in my experience because we would go to the customer. okay we've got this great agent framework.

its gotten identity. give some tools. he gave that goal. It'll try to apply.

it's given this identity and its goal in these tools will try work. a problem for you. But the customers are like what specific problems? well it's all for me.

i'm like let sit down together cup of coffee with brainstorm on this, can you sort again into like? by default your almost an fdm. not always you aren't after model where every customer spoke. And that thing you're selling for lack of better word is the ability to write agenting workflows Which is again valuable.

It helps people customers go live with production get value out of it. Where we're at least in terms of the last generation a third-generation? Is? I don't want to sell you a tool kit anymore, not only do You wanna buy one i'm gonna tell you an outcome.

so The current service now paradigm and its against more than IT case management. But I will go to a customer and they'll say hey You know act me incorporated. you did two million cases last year. I think i can have automation.

do ten percent of those and follow the same rules it'll use these exact same tools. its an anthropomorphic paradigm, you literally go to be user table create a user called a ipad. if u want apat to work case us sign up at and like. it follows everything you've already done.

And there is last mile. work on that paradigm. It's nowhere, no one has solved it where you literally just go in and flip a switch at ten percent automation or fifteen percent automation works but its way easier than going with an agentic tool kit trying to do the same thing. because we've done all of the plumbing.

There are buncha tools already here. know how look into your knowledge That sort of what. I think the puck is out. enterprise software like hey want take whole task which burning human being.

I made the whole task and i want to follow of a same rules that humans follow when they do it because frankly you should not trust in LLM or any trusted human being. And modern business processes were designed on the assumption that human beings are little bit untrustworthy. just follow the same rules as humans do like have an approver, have a manager who looks over the work having spending limit on an agent which is best spend money. so If you do all those things, I feel like that's a much faster path to value for customers.

And server-to-puck is going more generally out of the toolkit era and into this sort of outcome year. We want it just solve problems with our customers. Understood! To what degree has that rolled?

You come from traditional SaaS model where you basically charge per seat or... Oh, so specifically for that one we do charge receipt. We're a hybrid pricing model. you charge receipts But if you roll out the AI stuff most of it I stuff we charge per consumption.

Yeah. So from my point-of view like let's say i've sold you a thousand seats and your working cases And you rule up the AI? You know this is awesome. only need nine hundred people to do That.

I'm gonna have the extra People don't Do something else. For me The consumption side for the AI should at least offset the seats that you're not buying like, That's the way. The pricing model is constructed every year. point of view You should get better value out of it as well.

But that sort of a calculus Well I'd say. but there are two schools in the enterprise software space. There's one school thought which says i'm going to dig In and try To defend my traditional seat based licensing model by hooker by crook change make it really hard to orchestrate my stuff. Some people are there, that's not where we're at and We sort of feel like this is an automation revolution Like the whole point of AI has to automate work And were a Work Automation platform.

So whether It gets automated with human mediated or Not Human mediated? We want To Make it easy for our customers Which Is why you've got The hybrid price. When People Are Involved I Got Perceives. When AIs Involve Its Consumption That Balance Will Be Upto Every Customer.

Like there's gonna be customer specific and they're going to determine for themselves what the right model is. For them, so do you make sense? Yeah absolutely makes sense. and like just reflecting about it in a few years from now maybe a to-do list tool that we have on or I will have my Mac kind of contain like a Kanban board where i can assign stuff either by wife privately An agent, right?

And then an agent will review it. Or how do you see the future there?" I think your marriage must be different from mine. My experience is my wife can assign stuff for me but that's...

Right! That's sort of where the puck is going and I think in consumer space There are some attempts to bring this anthropomorphic model pretty strongly. I think the consumer space is sort of also dominated right now, but i'll call them interactive paradigms. It's like i'm gonna.

i'm going to chat with you which makes sense Like that's the obvious way. interact with AI The enterprise base. there are a lot of non-interactive workflows Or the interaction super limited? Like hey i'm working in a car factory and we're low on steel.

i need new steel. There's a huge purchasing system creaks into life behind the scenes, and it finds the right vendor in the right steal delivery time. And works with a railroad to get it in there. you're just that person at The Factory Center run alone on steel.

That purchasing system on the back end. It's not chat based. its may have some chats in their. that sort of orchestration stuff that like enterprises are like.

maybe I can orchestrate that whole process. i Can have an AI purchasing agent? Buy from the best vendor and some human being will approve it. And then, It'll just show up.

So that's a place where I think The fulfillment side is A more relevant piece in enterprise Whereas the consumer side Is more focused on the Initiation and the feedback loop. Like i asked for new car. When my car gonna Show Up? You know can I track it through the?

you Know maybe its interesting me to Track it Through the production chain See Where it is along way. And then back to your business model, or I mean that was about you're a business model. But what do think of valuations changing and the market being quite scared off by companies like Antrofiq? This year i think you lost forty percent value roughly Like have similar moments in your career already.

And then on February third and traffic ships at twenty dollar Claude Pluckin and Salesforce, you both dropped by seven percent a single day? Is that just nonsense or are Are you like how, I don't know. Afraid is the wrong word but right? How do we see that evolving?

also when you look at the real value that comes back from that and also you describe the value, right? You have agents who really physically do things. How does it relate to your model on? how do you see the future of SaaS?

or so me being super deep in SaaS Like...how does our future looks like? Yeah Well, I think in addition to my other good deeds. I'm obviously a pretty major shareholder and service now so personally goes up into the right.

yeah So you'd personally feel it when it goes down But me the underlying Market mark. Yes The market doesn't know what to make of it right? Now there's a hypothesis And you articulated it pretty well which says hey You know the fundamental seat-based SAS model It's being rendered obsolete by AI. Um, there's a strong version of hypothesis which says the underlying data repository in the workflows.

It's all irrelevant. You're just going to vibe code SAP In a weekend and it'll track your GL And you purchase orders. do all that fun stuff? There's a soft version Which is hey like That still has value but Like The front end abstraction like that interaction with the end users will take away from people Or orchestrated externally.

so we'll have little like barnacles on the battleship of a code base like steering this ship for you. I think they're both extremely overblown and i also thank that. be peace people miss is if u accepts the premise and actually do then just talking about it. but the future of enterprise software is less than last human mediated more technology.

It's more and more about automation doing the jobs. Who do you think has the best inside pole position track to provide that? Automation? it's the people who currently own the systems.

That's the People like this service. now is its work days, it's The SAPs. Will we all succeed? no I guarantee some of us will turf in.

But from my point of view Like if you look at things like our anthropomorphic or autonomous worker paradigm We can do that because we are the system that orchestrates. The workers, huh? That is very hard or in some cases impossible for a third party to do and so that Is the fundamental advantage. um did you know the current owners of these systems have?

but I feel like the market is So between undervaluing and misunderstanding. But what i always you know it told this too Couple people at our financial analyst day, and it's not secret everything. Financial analysts days public. But it's like hey I can say this to you And you cannot along or you cannot.

but so you agree with me? Oh yeah makes perfect sense Like the proof point is when it shows up in our numbers. Um and fundamentally like that's the currency of trust. When you're going through one of these paradigm shifts Is like It's nice to be able to have a coherent narrative into believe something But to want two things a third party you've got have the revenue.

You gotta have the customers and at least for service. now i would point to our numbers. this is that frustrates bill. we been having great quarters.

see what the next one looks like. but if we published them all, and so I think it's your investor relations people or bill they're like yeah this is overblown uh... like the way to get out of that bucket is partly by Talking like this, but it's also partly going to come down too. We can have a couple of blowout corners and point them be like hey You should pay more attention to us now.

So that's one of our goals. we want to have awesome technology But we got to bring into customers too. so Like an uber positive you could basically say That AI is a huge chance for you Because your own the system off record To basically offer service as an add-on to many, many customers really use your software and orchestration already. It is in a way like massive upsell where I would say it's maybe worth more than the database that you've built right?

The immediate To offer this as a service to so many customers. It gets absolutely right i think. the high enough level of abstraction. I mean, we're workflow automation company.

yeah We help you automate work and traditionally we did that with the assumption being... we would make it more efficient for human beings to work. A process through an enterprise will track things forward than they do. The handoffs who do the assignments gives us in new tool.

We are able to automate some stuff we couldn't beforehand, or we can automate stuff like much more efficiently. And that means that we can give our customers... Like we're even more valuable to our customers. and does I think gives us the sort of inside track on this?

There's a soft piece too which is subtle in the...we know this space-like. we understand enterprise workflows And we just fundamentally know what purchasing workflow works like or an IT workflow, a customer service work flow. Which I think does give us some domain expertise which is not always present in others.

Okay okay then let's pivot back to nerdy stuff even though everything is AI these days. If you think about the core value of software and maybe some things like pivoting more towards services that you offer in addition, then the software development lifecycle is also fundamentally changing. And how do you see this? I mean, we basically move from the bottle neck on the engineers to like a bottleneck on product.

To potentially even a bottlenecks of your customers right? Like i mean... Product life cycles or product work is also validation with our customer and product discovery Right! Like hey client XYZi want you show something..

I made this great tool And I myself, I observe from my own behavior that I'm slightly tired by new software hitting me every day. Like there's so much new software popping up like out of the blue right now how do you see that changing? Is their natural slowdown or a natural bottleneck in this process which is now moved from engineering to not only product Which many people say about the customer? And are we slowed down by the pace of our customers?

So is that maybe like a natural break for, for... For the whole market. Um it's just a really good question and I'll actually throw us a third one if you didn't mention. but its real too at least in enterprise space.

Like i can iterate super fasts and the customers have got their own rate of willingness to change. In middle You've Got like The go-to-market and the field and the services organization who Are talking To the customer interface. they gotta get The maximum rate at which I can push chance through there as well. Um, my instinct is that yeah over time we will slow down because you saw this in early sassas.

Well like um early days of service. now i literally did nightly pushes to the customer base And then went a weekly and they were going to monthly. and it's seasonal. It's a combination of two things.

One, the customers are like hey I've got more and more mission-critical workloads on you... I can't afford to have it. just change behaviors. even if nine out at ten times it got better The one that is ten times that I as a customer thought was worse.

That's a problem. And second thing is the products just get a better fit. There was less than need to chase the feature set in their design. So those two kind of converge.

at some point You get slowing down the iteration rate. What we in service now are trying to do is I want to be able the iterate the underlying engines as rapidly as i possibly can because if you like. Hey look if your assigning cases to a ipad he's supposed work on and right now he can do twelve percent of the case is, if he just does fifteen percent other cases that's better. you don't change anything.

it should make sure happy. If ipads behavior changes though like periodically he's gonna help them Start asking a i married to assist in the cases. look back and forth. that changes your process.

I feel like you need some amount of control as to when i marry joined your team. so is this service improvement without changing the flow, what does it case where i think the modern paradigm is just want to go fast possibly can. we're forcing change on users. enterprise space like Everybody's a little afraid of being like snapchat and you roll out the new UI there was.

I don't like it. Um, so in enterprise space There is more of A with customers need to have control Of functionality changes that impact their end users. But even without controlling place You know i used to release code twice a year. now It's like Customers are getting pinged probably weekly or certainly monthly.

So I gave us a new version. would you like to turn it on? Yeah, yeah and customers will start ignoring that always On green lights... I got some of her like hey stop nagging me.

hundred percent Even with really big corporate entities. so think man You're gonna be really conservative who are? no all the things That i was hoping for now here right It's like, hey that new thing. That thing I was really waiting for is here.

let's go cool. so it's yeah the we're looking forward to hear. i would say software was ever a boring industry but it was predictable from maybe ten years here he does less predictable. now if you are living through a technology generational shift on technology and its reshuffling the deck unlike what companies are necessary paradigm succeed And it's resurfing deck on the customer side too like there.

they're not sure how their supposed to behave as a consumer. Um and so they are feeling away. same is the rest of us in. some of them want to go faster someone wanna go slower, some other afraid some other super enthusiastic uh...

and probably take out you know give it few more years, but right now it's a transitional era for sure. Let us give until we add robots! Right? That is the whole other space like.

I mean. there are hundred percent people as you know this who were doing physical AI and supplying slightly different LLMs to physical world tasks. That's gonna be crazy, so I only have two questions left for you as time is short. One and that's like beginning of the outro.

So for the CTOs listening most of them Running like slightly smaller shops than yours What would you tell them about wiring agents into their real systems this year? Like What to do and what? just stop wasting time on? yeah, so I guess give two different pieces of advice.

You know one there's an aphorism. I think that like everyone is conservative about the thing they know best And like the thing? I know best as engineering Like if you're a CTO but then you probably no bests also engineering. This is really not a time-to be overly conservative.

The new tools are out there. they will absolutely help your team be more productive. You can get more product out to your customers faster by leaning into it, and the flip side of that is unless you're an incredibly rare nook-of-the-software industry... The best way you could make YOUR customers more productive - give them more value - is in making your products smarter… And there's a whole raft of things now that YOU couldn't do three years ago!

Um, so like if you're not using AI to make your own internal teams more productive. You behind the curve and my advice would be like it's not too late just like now is a time like lean in catch up um talk To Your peers because some people are there Some People aren't. And The second Thing Is If You're Not Applying That Ai To The Products You Sell? In Doing So An Intelligent Way You're Gonna Get Lapped By Some Other Company that Is doing so So that i imperative is it a hundred percent real.

It's on the internal and the customer facing side of it, um And I guarantee some of us will turf it. we'll like put some bad in our product which is annoying or terrible. but The flipside to that is like not making that investment is just surfing at different way Because you know two years from now somebody else with figured out. so this isn't time for excessive caution.

It's not a time to completely go bonkers and do crazy stuff, but this is the time really lean into that new technology. The foremost real yeah. So thanks a lot Pat. I still have less surprise for you.

so someone on your team told me about a hidden Easter egg in? Now platform yo workflow engine. if you open an inch incident That doesn't exist and hit say three times Then there's little like a feature that makes you scroll back in the timeline and you can stop it. In the year, you want to stop then?

And assign a name which we now assigned your name. We travel back to Fred Lattie days... You said like two thousand five right roughly before ServiceNow was really ServiceNow. And yeah, you get to watch your younger self now for a bit like coding.

Like building your own date functions in Java right? The good old days and have the chance to whisper something into young pet ears what would it be? Oh man I'll give. on the technology front i think we actually made lot of really good decisions On the tech stack We did.

But I would whisper in my Odears like, hey that UI rendering layer you're using. Like that jelly XML templating language it's on the top of Apache is not going to stay there very long. You should maybe look at a different UI templateing language or like onto third generation since then but its still. they are buried and the guts of system.

But no longer XML based? Or...I hope! There actually That we are a third generation of ui rendering.

the first generation is still rendering some your stuff and it's absolutely its tag based xml transformation. I think on the, i would also say second piece that they will be enjoy. it was incredibly hard work And I remember being incredibly stressed because i have this huge pride of ownership and like, want every customer to succeed. everything i checked into be perfect every customer around well.

But i think i never really came up for error. instead like the the positives about environment like that the closeness i felt with the team the validation i got four like customer has a question and i go back to my hotel room and i work all night check it in shows up build the next morning shop next morning show. that's amazing. this is what I wanted.

Like those are rare in life, um and i think just the advice you do enjoy that. on a third one more personally i would probably tell myself hey getting all this amazing validation out of uh... work and that's great but don't forget to spend some time with your family Like your when the rare times you're not working, that temptation may be to retreat to your room and go to introvert cave. And you know, edge out but like Your family wants or needs whatever The right word is.

But like in you need them. So factor that into the way You're spending your time as well and spend more Time there. make it more valuable time When you spent. Thanks a lot Pat.

really great having you on the podcast Really liked a lot of helpful advice. And yeah, like also I have way better understanding of service now and what it is how to functions internally. thanks a lot. hope you see soon maybe in real life.

at a certain point You're in Santa Clara right? So i'm San Diego but reciprocally It's a pleasure! I appreciate the interview. Yeah happy to chat another time as it comes up.

Thanks a lot. Have great day Bye. Thank you for listening to the AlphaList podcast. If you liked this episode, share it with friends - I'm sure they love it too!

Make sure to subscribe so that you can hear deep insights into technical leadership and technology trends as they become available. Also please tell us if there is a topic you would like more about or a technical leader whose brain you'd want us pick? Please send us suggestions to cto at alphalist.com, Send me a message on LinkedIn or Twitter.

After all the more knowledge we bring to CTOs The more growth We see in tech Or as we say an Alphalists Accumulated Knowledge To Accelerate Growth. See you In next episode.