Pendo CTO Erik Troan on the Rise of Vibecoding in the Enterprise

Hello and welcome to the latest edition of the Techstrong AI video series. I'm your host, Mike Zu. Today we're talking with Eric Choan, who's the CTO for Pendo, and we're talking about vibe coding and the impact that that's gonna have on enterprise it.

Hey Eric, welcome to show. Thank you very much for having me, Michael, For those that don't exactly know what Vibe coding means and is, maybe you could describe this a little bit 'cause I think some folks listen to this and they're like, well this must be something left over from the sixties. Yeah, I mean, I'll, first of all, I'll give you the very extreme vibe co coding definition.

Now I think the term is all of two months old or 10 weeks old, so it's pretty new. But at the extreme it means using an AI to kind of do everything for you using human language to drive the ai, even to the point of just doing really rapid iterations. So you tell the AI to build something, it's not quite right.

You tell it to fix it, it's not quite right and you just keep feeding the AI and give, going through a really rapid set of cycles with it to try and get something that looks like what you want. It's not test heavy, it's not spec heavy. It's not like people are writing a whole bunch of cars in Jira and moving them through a lifecycle to define everything.

It's just sitting down and trying to get an AI to do something for you and get some working code that you can go use for your next step. To your point, it hasn't been around very long, but it sure has engendered a lot of love and hate on both sides. So yeah, Absolutely.

Um, there are folks out there who say, this is great. It means mere mortals will be able to build software and others that say, well, the problem is mere mortals will build bad software. So, um, how do we kind of think about this and what is the impact that that might have on the way we build applications in the enterprise?

Yeah, so first of all, I say why not both? Like, I think both of those things can absolutely be true. And one of my favorite things on vibrating vibe coding came out was this Twitter thread from somebody saying, oh, this is great.

I did this. I'm bringing it all to productions. This is amazing.

I didn't have to hire any engineers. The next day it's not scaling. Oh my God, what's going on the next day?

People are attacking me and breaking in. Third day, I'm, I have to turn this off now. Why can't we have nice things?

Right? Um, and that's certainly one extreme of it. Um, that's probably the one where people are saying, oh, this can't work.

The other one is we have people here, app penda who are non-technical. They might be designers, they might be product managers, they might be salespeople and they're using vibe coding to go bring an idea to life except they're showing us what they, what they're thinking. It's a way of, uh, doing more than a storyboard and more than a picture on a whiteboard.

They can generate something that looks and feels like working software. So it's almost prototyping writ large and as a way of communicating ideas and thoughts there. It's a, it's incredible.

There's nothing really like it and there's nothing for that speed. So that's on the other side, right? Well, where's the reality gonna be for the enterprise?

It's probably gonna depend a lot on not only your risk tolerance, but your, how, how important is it? It's correct. Look, I have money in a bank.

We all have money in a bank. Do I want someone vibe coding the application that's transferring my money into my kids' allowance every week? Not really.

Like that would make me nervous. Do I wanna vibe code some workflows at Pendo that we're using with our BDRs in order to have them identify leads, scrape information off LinkedIn and figure out who the right people to target are and who the people would be receptive to Pendo um, Pendo message. Yeah, sure, let's do it.

Right? There's very different cost of failure, very different testing requirements, uh, in the sense there's not only even the latter one, there's not even a correct answer While in the bank case there is a correct answer and it's really important you get it. Will this kind of change the way we interact with developers and the people that they support?

'cause historically, and I think we've all been there, people come up with an idea, they put something together that feels like a requirements document that gets handed off to somebody, it comes back, builds something, and uh, typically it doesn't quite match what the original vision of the thing was. And then it goes back and forth to half a dozen times till everybody gets really frustrated. And typically what we wind up with is a shadow of the original idea anyway And we didn't give up.

So with vibe coating, can we get to something that does feel a little more interactive and kind of close that gap between, you know, what some people would say are left brain and right brain people? Well, I think it's gonna definitely close the gap, but I also think it's gonna re uh, create a lot more software that's getting deployed. So I think you're gonna find non, um, software specialists building solutions and expecting 'em to run.

Look, I mean the cloud was, if you go back 20 years, the first step in this, all of a sudden people could start deploying things without having to go through it. Now people can deploy more complicated workflows, more complicated agents, even more complicated UIs without having to talk to it. And they're gonna do it like it's crazy for us to sit here and think we're gonna be able to control that.

We couldn't control cloud because infrastructure as a service, we couldn't con um, control business SaaS applications. This is going to happen. So I think IT organizations need to understand what does it mean when their employees are vibe coding things on the weekend and getting anything that helps 'em do their job?

'cause the whole goal is from it is to help people get more productive. I also think what you said is true. We're seeing App Pendo where our designers are vibe coding things and bringing 'em to engineers and the fidelity of comm idea communication in a vibe coded application and a rich prototyped application, is it an order of magnitude better than what you could build with Figma, which was an order of magnitude better than you could put in a Microsoft doc.

So you'll have much better cycles, much more rapid cycles. It'll also either allow or force product managers and product designers to think through edge cases a little bit, right? I've said that one of the skills of an engineer is finding places where maybe the spec has gaps in it, they'll see the gaps in real life and they'll get to go fill them in.

Now that's a case where I think a lot of those vibe coded apps are gonna end up being rich prototypes and where engineers are gonna have to rebuild them for industrial use, at least for a while. But I think it's gonna change a communication dynamic significantly. So what happens to all those low-code no-code tools that we had for all these years where we were trying to create this, uh, community of citizen developers with mixed success?

I mean, do they go away or are they gonna be more just vibe coding platforms? I think they're gonna become vibe coding platforms or they'll go away. I think those are the only two choices, right?

There's a reason that Python is such a popular vibe, coding language, it's easy to learn, it has no memory management. You don't have to be an expert on how computers work on the inside. And a lot of these low code, no code solutions have these scripting languages.

You don't need a scripting language in a, in a world of AI and gen ai, you need human language definitions of what you want to do in an LLM to turn it into something more rigorous, which is really what vibe coding is. So I think you're gonna see vibe coding certainly generate JavaScript code like you see it now, but platforms like Bolt, it's generating something. I don't even know what it is.

Nobody cares, right? It's just, it's being some something they can execute my idea and frankly a lot of agents are an instance of vibe coding. Like if I wanna build my own agent that Rafis personifies something, I'm doing it with LLM instructions, I'm doing it with sample documentation, I'm just throwing stuff at the wall.

How that gets executed, I neither know nor care. It's about the business value. It's about building an application or an agent that people can interact, uh, interact with and use.

And it's interesting to me is AI agents are an instance of vibe coding. Well, um, we're using them to create code written in Java and whatever else it may be, which we created as an abstract to provide an interface between humans and the machines. But if the AI is a machine, maybe it will write something in a programming language that's a little more efficient than the one that we used for humans, right?

Yeah. I think there's two possibilities that are really interesting and neither one may come to pass. One is what you said, why are AI writing Python code instead of machine code?

Like it's not entirely clear to me. Now the reason is they're not good enough at writing machine code, right? And people have to, humans have to read it and fix it right now, but that's not gonna last forever.

We have the worst AI today we'll ever have. They're only gonna get better. Um, so that's one thing that's interesting.

And another one is where is the line where an LLM is just your execution engine as well? Like in an agent you're generally writing English code and the LLM is the execution. It doesn't turn it into Java code, it doesn't turn it into Python, it just goes into executes your instructions.

And an agent is doing it in a loop, right? It's going and repeating tasks. It's doing things for you without a lot of code being generated at all.

So is the future end state of all this i human instruction to generate Python instructions that get generate machine instruction that seems a little clunky, right? It does seem like there's an extra step or two in there. I also wonder if it will change the way people work in the enterprise because it seems to me it will become a lot easier for somebody to say, here's my idea, here's my idea in software, and then share it with other people who aren't developers and they will collectively build something whether they build on top of each other because a lot of that innovation is still, uh, you know, a group does better than a single individual most of the time, at least for the long haul.

So will the way we kinda work together change? Yeah, I mean, individual software use software development used to be a proprietary exercise, a solo exercise where you went in the closet and did it right? You think back to the movies about 1960s, um, MIT closets and now it's a very collaborative effort.

These first vibe coding tools, they don't have collaboration, they don't have version control, they don't have the ability to work together very well yet, but it's gonna happen very, very quickly. You're gonna find the, uh, uh, the hugging platform or the GitHub platform. You have one of those things for vibe coding tools where people can't collaborate and work together very rapidly.

Now, professional developers are usually supported by something that feels like a DevOps slash software engineering team. Will those folks be then working with a lot of the applications that are created now by citizen developers who are vibe coding things? And will that change the way that that whole process works?

I mean, your guess is as good as mine. My guess is my guess though is that you're gonna find those teams building a platform that these applications could be run on without taking responsibility for those platforms or for, without taking responsibility for those applications. So I'll say, all right, you want to use this vibe coding tool at my co at your company?

Great, you can use it. Here's where you run it. Here's the cloud environment.

We maintain the cloud, we give you the ways to deploy and run your agent or your whatever it is that you just built. But those people, the people who wrote them are, are gonna have to take responsibility for 'em. I don't think any IT organization's gonna be able to take five applications a day that are being spit out across an organization, a scale of 1500 applications a year and have any meaningful ownership of it.

But will they provide the core database that's used? Absolutely. The will they provide, uh, 24 7 infrastructure to run it on, you know, definitely.

So they'll provide the enablement and the platform, uh, layer of things and then they'll just see, you'll see this huge number of applications just running around. On top of that, One of the ironies though is when we tried to have this citizen developer revolution using low code and no code tools, uh, at least in my experience, half of the people using those tools were actually professional developers who just preferred to use something at a higher level of abstraction or write something that, you know, they kind of just need it quickly. Is that gonna happen all over again?

And engineers want to go fast. They want to build software quickly. And the story of a kind of innovation as we went from assembly to low languages like c to higher level languages like c plus plus to Python and Java, at every one of those movements you've seen engineers climb up the stack to something that let them do their job faster.

Vibe coding tools are gonna be another layer that lets them do their jobs faster. And I think you see that in data scientists probably first of all where they spent the last 10 years learning Python or data science libraries. I like, I have a kid who wants to do data science and I'm saying, why would you learn to code, like just go use some platform.

How you put to get through Slink Python together, do matrix operations is not interesting. It's understanding what matrix operations need to be put together. And these tools will get you there faster.

And that will happen across all the whole development lifecycle. Will everybody eventually start to think more like a software developer? Because there is, it is not just the tools, it's the way you can think about problem solving.

And maybe, you know, we collectively as humans will be trained by the AI to think differently. No, I don't think so. I think that's actually exactly the opposite, that all these tools are about letting people think like humans in the machine that we able to adapt to human think so.

So it's, we'll we'll see. By the way, I think that's interesting, but I don't know that everybody in the world needs to be able to algorithmically decompose problems to be successful with in a AI vibe coding world. So what is your best advice then to CIOs who are looking at all this right now with probably a mixture of odd anticipation and a little fear?

I think all this feelings are probably like the ones they had when mobile happened. Probably like the ones when web browsers happened and the internet happened. Like this isn't anything new.

This is the next seed change. What, what did we learn from those? That if you try to stop it, you're gonna get run over.

If you try and prevent cloud, everybody's just gonna deploy cloud and you won't have any control at all. So be open to it. Enable your your team to try things, enable experimentation across your enterprises as broadly as you can.

Literally today I tried three different coding, AI coding platforms to solve the same problem to see which one worked best. I won't name them and I won't tell you which one worked best by the way. Um, but we're try, that's what we're doing here at Pendo.

We're getting encouraging our engineers to try all the tools they can. They find try everything. Make sure that you're not only building your individual skills because you know, no job is forever, but you're helping Pen to learn which are the ones we should embrace.

Now, my ciso, he's absolutely approving all the platforms. We are looking enterprise deals with all the different providers so that we can use single service si single service sign on so we can have the right levels of accountability, the right levels of auditability. So we're trying to let people use the tools they want to inside of a IT compatible universe.

But that can't mean just saying, no, you can't do it. 'cause then they're just gonna do it on their personal credit cards. It's true.

We'll just have another flavor of shadow it, right? Yeah. And one that is probably 10 times worse, right?

It's that enough. When you use random applications, can you imagine writing random applications? You need to find a way to have corporate control, corporate blessing of it.

Also, will the applications themselves become a little more disposable? And I asked this question because so many folks have an idea, but by the time they stand up the development environment and do all those things, they just decide to go back and sit on the couch. So will we have, you know, things that people are gonna do that they only need for, I don't know, a month If it only takes you, if it only takes you an hour and a half to build?

Definitely. Mm-hmm. And you see, ID you see computer specialists doing this all the time, or how often do we throw together something new script?

We use it for one task and we throw it away. And we do it where our peers don't because we can do it quickly, but people do it in spreadsheets. Who are spreadsheet experts?

Like is when the, when the difficulty and the expertise you need to automate something, you know, goes down the number of things you automate goes up. It doesn't mean you need to automate it for the next 10 years. You may only have to automate it for the next two weeks.

Right? And it's one of the, it's one of the places where I think you can see the most productivity improvements from this, by the way, is everybody being able to automate little, little tasks as they need, as they see fit. All right folks, you heard in here, there's no avoiding it.

And frankly, if you're an IT leader and you want to hang out with the cool kids, maybe it's time to host that vibe coding party. Hey Eric, thanks for being on the show. Thank you, Matt for having me.

Enjoyed it. All right, and thank you all for watching the latest episode of the Techstrong AI series. You can find this in other episodes on our website.

We invite you to check them all out. Until then, we'll see you next time.