The State of Platform Engineering Today: Pankaj Gupta on Why Platforms Hit Their Ceiling | Platform Engineering 2.0 Ep 1

Hey everyone, it's Alan Shimel from Techstrong. I want to start right off by introducing you to my friend, Pankaj Gupta. Pankaj is with Broadcom.

Truth be told, if you've watched me and Pankaj on video before, we know each other for a very long time, well before Broadcom. Well, not before Broadcom, but before Pankaj was at Broadcom. 0.

Each episode's only about 10 minutes long. It's a series of questions and conversations. I hope you're going to enjoy it.

There'll be different formats for different social media platforms and wherever you consume videos. tv. You'll be able to get them there, or our TechstrongTV YouTube channel.

Anyway, Pankaj, welcome to TechstrongTV. It's great to have you here. Thank you.

I'm very pleased to be with you again. My pleasure. So, you know the thing about episodes one, you're always going to do a little housekeeping, always going to do a little foundation building.

And there's no way of getting around it. 0, a refresh, if you will. And that's what we're going to try to answer in this particular episode, and in the future episodes, we're going to delve in deeper.

Before we start, I feel like people know me because they're on TechstrongTV, but they may not know you. So if you wouldn't mind, give them a brief little, your background, your journey. I'm part of the Broadcom's VMware Cloud Foundation division, which is the full stack for build the private cloud.

I'm senior director of cloud solutions at Broadcom. Previously, I worked with VMware, and before that, a few startups, Citrix, as well as for Cisco for a very long time. My passion is about helping the practitioners as well as executives about shaping their IT for the next evolution.

Love it. And I think you were still at Citrix, I think, when we first met. Yes, I was.

Mm-hmm. Let's not say how long ago it was. We'll both be happier.

0. 0. But in our conversation, I think this is a good place to start.

Platform engineering burst on the scene, I don't know, five years ago, four years ago maybe. It's not that it was necessarily new. A lot of the underlying pieces of it we call ops, and a lot of people knew of it, but the packaging perhaps was new.

When it first started, it was more about managing Kubernetes because frankly, Kubernetes was that hard, right? And you needed something to do it. But what has fundamentally changed, in your mind?

Because over the four or five years, platform engineering has become a real thing. It's widely, widely adopted. Yeah.

Right? And you are absolutely right that platform engineering is no more optional, and adoption is almost near universal, but maturity varies across the organization. But if you look at the platform engineering, it has been a big force in the marketplace and a standard practice, but has one single goal was to remove the cognitive load on application developers, increase their efficiency.

There are three primary components if we summarize for platform engineering. One was it standardized the process to take the course to the production through the golden path, and that has been the huge transformation for code to production for any organization. Second one is the internal development platform, so where the developers move from opening the continuous stream of tickets to click ops click and the environment builds for that, and it reduced the cognitive load on the developer.

It also made a lot of automation for platform engineering or IT organization. And the third one, which I see is more opportunistic because once you have pipeline, you can standardize putting the code scanning into the pipeline, and that's where the shift left came into that. And though all three are the great foundation principles for platform engineering and companies start managing platform as a product for that.

Of course, the maturity varies for that. But the last two years, there is a huge transformation is happening in the marketplace for that, and I will say a revolution which is driving the evolution of the platform for that. Let's look at first one is that, first one is the AI-driven coding acceleration.

I think every developer is using AI tools for that, for writing the code for that. The organizations who did four releases in a year, they can do 40 releases in a year. But there is a very strategic implication of that.

The strategic implication is that the bottleneck is shifting from code generation to taking to the production for that. Pipelines were not designed for that. Second big thing also happened for the role of a developer is also changing from writing the code to verify, validating and taking to the production.

Second big fundamental shift, which we see in the marketplace, is the agentic future for that. There will be more agents than the developers in the world today pretty soon for that. And even Satya Nadella has said things like that, the number of agents the Microsoft will have one day into that.

So those are the, when you combine the agentic future and the impact of the AI-driven coding acceleration, that itself is a huge transformation for that. Yeah. There is also additional impacts which AI brings into that, the need for the FinOps, which is stronger.

We have seen from token maxing to the token economics. The two new words just popped up in the last two years for that. Yeah.

And agents are also influencing very big part of the security and the AI is widening the security gap more and more for that. And there has to be checks and balances. So if you see that AI is the nuclei or the center of driving the revolution in the industry, and that require platform to evolve for that.

The platform is hitting the ceiling for that. Fair. You know what?

You actually hit my second category, which is what are the forces pushing the platforms to this breaking point? I think you did a great job of explaining them there. But putting that to the side for a second, let's discuss where has sort of the original platform engineering platforms most commonly hit their ceiling?

So not necessarily what's new even, but even with what was old, if you will. Yeah. They've sort of ceilinged out.

I think there are a couple of ceilings. If you look at the drivers or the forces, the number one is the AI-blind architecture. Even if you look at Kubernetes today, they don't have visibility of GPUs.

Forget about the platform itself that. So platform has to support AI, and there is specific needs for that, like the GPU provisioning, pretty soon the TPU provisioning for that, need for models, serving the models for that, MCP server. So platform has to integrate all different new requirements for the AI acceleration for that.

The second ceiling which is hitting is pretty soon or actually now is that evolution from developer-only focus to multiple personas. The data scientist is one of the big things. The second one is the FinOps for tokenomics to the-- Also that the previous FinOps look like more bolt-on reactive rather than the proactive for that.

Security was good consideration on the shift left, but security gap is widening, so security has to go deep into the platform for that. And more I think about it, golden paths have been golden for platform engineering. But in last two years, you have seen that every CIO, every chief AI officer in many organizations, there is a lot of experimentation has to be done.

So golden paths are really golden for a standard workloads or standard production. But when you have to do a lot of experimentation, new models, new mechanisms, they feel like a little bit of handcuffs to some of the organizations for that. Regulatory environment, if you look at in Europe, they are evolving continuously.

If you also look at the regulatory environment for AI in the United States is evolving very significantly. So the golden paths are becoming golden cages for some environments, but not all of them. It's still fundamentally good for 80% of the environment.

But when you have to do the fast experimentation, fast learning, fast fail, fast success, that's where it starts becoming more rigid and more aesthetic platform for that. So those are some of the ceilings which we are hitting. You, I, and our friend Luca Galante were on an episode of the "Platform Engineering Show" podcast last week, and we talked about evolution versus revolution in technology.

Yeah. Very little in technology is truly, truly revolutionary. It's usually evolutionary.

There's a revolution, and then the evolution builds. Yes. 0 evolution or revolution?

It is absolutely evolution. No reset over here for that. Every IT organization, every C-suite hates, what do you call it, reset or rip and replaces.

Absolutely not. 0. They all carry over.

They become actually more important. Managing- Yeah ... the platform as a product.

0 underneath it? And I think the answer is no, right? Absolutely not.

You should build on top of. And so therefore, it can't be revolutionary. To me, that's the very definition of evolutionary, right?

I think market forces can be revolutionary, but platform will be evolutionary. You got it. Last question.

Well, two questions in here that I want to finish off our first part. 0 is AI success, right? We're all grappling with the token maxing, with where do you use human in the loop, where do you use AI?

How many agents, people, all of that. Is platform engineering, in your mind now, the critical path, the golden path, if I could borrow a phrase, to enterprise AI success? Absolutely.

Yeah. No doubt there. I agree with you.

And there'll be people who say, "I could do it without a platform," but I think those are going to be outliers, Pankaj. 0 is the process efficiency, standard processes, measuring the success. 0.

Absolutely. Last point we want to talk about in this first episode. 0 framework, and by the way, it's in the white paper, if anyone goes to download the white paper.

But for people who are too lazy or don't want to click through, give us that framework in a nutshell. First of all, it is, repeat again, it's the evolution, not the revolution. There are five fundamental pillars directly along what market is really asking.

Number one is AI native platform. Number two is multi-persona experience beyond just the developer. Number three is embedded FinOps.

You talked about tokenomics and the stronger FinOps requirement. Fourth is security shifts down. And number fifth is the composable by design.

Excellent. 0 paper. Go check it out.

But they're in here. We actually spoke about it on our podcast with Luca as well last time. Pankaj, thank you for joining me here for episode one.

0, AI native platforms, et cetera. Stay tuned. This is only the first of a six-part series, and we hope you'll catch the next ones.

But for now, I'm Alan Shimel. On behalf of Pankaj Gupta, Broadcom, myself, and Techstrong, thanks for watching. Thank you.