In this RedMonk conversation, Pankaj Gupta, Senior Director of Product Marketing, VCF Division, at Broadcom, chats Platform Engineering in the AI era with Stephen O’Grady. Gupta explains that more than 90% of organizations run an internal platform, but despite widespread adoption, Gupta thinks most platforms are hitting two ceilings. They were built, in his phrase, “AI blind,” and they were built for exactly one user, the developer. AI has already moved the bottleneck from writing code to getting it into production, and the user base is about to change shape. Also in this episode: when golden paths harden into golden cages, why FinOps and security can no longer be bolted on, and Gupta’s five pillars for platform engineering 2.0.
The RedMonk conversation is sponsored by Broadcom.
Links
Transcript
Stephen O’Grady (00:09)
Good morning, good afternoon, good evening. I am Stephen O’Grady I am the co
founder of RedMonk, the developer focused industry analyst firm. And I am here today with Pankaj. Pankaj, would you care to introduce yourself?
Pankaj Gupta (00:20)
Thank you, Steve, for having me on your show. I’m Pankaj Gupta. I’m part of the Broadcom VCF division,
Stephen O’Grady (00:27)
Okay. Well, that is a perfect introduction since we are here today to talk about platform engineering and sort of what the what the future is for the platform. So as we know, like there there’s a we’ll get into this obviously over the course of the conversation, but there’s a lot going on with platform engineering. you know, any number of trends, you know, sort of our impact again. So what is your just to sort of open, you know, with a broad strokes question, what’s your what’s your sort of take on the current state of platform engineering today?
Pankaj Gupta (00:54)
The platform engineering is no longer optional. More and more organizations are deploying or have built a platform and have invested into platform engineering teams for that. If you look at the reports from Google DORA report, which came out last year and December last year, we sponsored as Broadcom a
State of Platform Engineering report with platformengineering.org, all the data suggests that more than 90 % of organizations have adopted platform engineering and adoption is continue to increase significantly. Now the AI workloads are coming into that. However, few challenges still remain ahead of us. Adoption is good.
but the maturity of platform engineering and use of platform by the developers varies from organization to organization. It’s still a very significant road to go. Another thing what platform engineering teams have done, they have sold the value or delivered or articulated the value to their IT and business leadership also for that. One of the major…
to one of the way you see the success of platform engineering is when you start measuring the success of platform engineering. And the data suggests from 70 to 80 % of organizations have mechanisms to measure the success of platform, whether it’s a DORA metric or developer productivity or how fast they can take the code to the production, variety of that. So in nutshell,
Platform engineering is really living to the promise of delivering the reducing the cognitive load on the application developer, making things much more efficient, bringing the golden path standardization across multiple engineering organizations for that. Also improving the developer experience, improving the developer productivity.
moving from the ticket ops to the click ops for that, offering self-service portal for developers for that. also, platform engineering has also become a substrate on which you can implement shift left because now you have the standard pipeline CI/CD, so you integrate CI/CD. So it’s a very multifaceted.
function. It is complex. Adoption is getting more universal, but challenges still remain ahead for that. Many advanced organizations have started managing platform as a product. So when you start managing the platform, you start treating developer as your customers. What is their need? What is their requirement for that? So we are pretty excited about the platform engineering evolution, and I think it is at the fork.
roads today in the context of AI.
Stephen O’Grady (04:23)
Okay. So, you know, when when we have conversations with platform teams, and as you say, you know, i i if they’re not ubiquitous, they are sort of becoming close to that, right? It’s just a sort of necessary component of of sort of any technical organization. you know, we see them sort of hitting, you know, either either hitting or approaching a ceiling. you know, do you see that and sort of in your experience, you know, sort of why is that happening?
Pankaj Gupta (04:50)
I think last two or three years have been big changes in the industry as you know for that. Originally when platform engineering started, it was AI blind architecture. And platform is becoming a substrate to build, deploy and govern the AI based applications for that.
So that’s the one ceiling some of the platforms have started hitting, not being the AI aware architecture or not able to enable. But the good news in the last two years, many platforms like VCF and others have crossed the red boundary of not being AI-blind architecture. So that’s the one big ceiling is hitting. Second big ceiling, which
current platform has to evolve or go beyond its ceiling is the developer only focus. When the platform engineering started, the primary focus or primary persona was the developer. Now, there are variety of new personas who will use and have a benefit from the platform. First and foremost is the data scientists. Next one will be the FinOps.
the security teams. So platform engineering has to evolve to support multiple personas for that and not just the human personas, now the agents also come into that. And I think platform engineering has been mostly reactive instead of proactive, especially on few areas.
One of them is of course the FinOps. You know what is spent has happened in the rear view mirror. We should look at how is spent is going to be in the future or with the new environment for that. Security has been a bolt on security. Security has been, or the compliance has been more rear view mirror rather than the forward view mirror into that. And there is another fundamental shift which is happening for certain use cases, especially for AI. The golden paths.
are excellent and will remain a foundation for platform engineering. But in the last two years, lot of experimentation and learning is happening for AI, and every six months or every two weeks, there is something new comes out. So you need lot of flexibility, experimentation, fast fail, fast learning. So golden paths are good for a standard application, but when you’re learning significantly,
Sometimes golden paths start feeling like the golden cages. So you need a lot of flexibility into that. The current platform engineering or the original platform engineering were more focused on rigid and static platform. If you talk to me last year, we would have not talked about MCP servers. We would have not talked about the AI gateway. And then just
Stephen O’Grady (07:56)
Right. Right.
Pankaj Gupta (07:58)
all the new building blocks coming up and their choices of that. So those are some of the ceilings for that.
But many of platforms have already started moving to the platform engineering 2.0 and that’s the reason Broadcom worked. So shaping the narrative or shaping the way our platform engineering has to go next as a platform engineering 2.0. But I will reinforce that the next iteration of platform engineering is not a reset. It’s an evolution. It builds on the fundamental principles of platform engineering.
improving the productivity, standardization, reducing the cognitive load for that, managing platform as a product. So all those fundamental principles of platform engineering still remain extremely valid and will be there for a long
Stephen O’Grady (08:49)
And so for the next question, you’ve already touched on this. you know, obviously everything in this industry at this point in time is being
Pankaj Gupta (08:58)
Yes.
Stephen O’Grady (08:58)
impacted by AI. and platform engineering is no exception. so in in your view, where does you know, where do AI and and platform engineering intersect?
Pankaj Gupta (09:10)
I think.
It’s a very strong intersection and timing is absolutely right for that. And let me go a little bit more deeper. If you talk to any developer today, most of the developers are using the AI for writing the code. So there is a significant AI-driven coding acceleration for that. And independent of how big organization is the developer or how many developers are there for
More and more developers are using AI tools, and AI tools are extremely effective, highly productive for that. But it has two major implications for that. Previous bottleneck was that writing the code
Now writing code is becoming like a commodity or you can create much more code. So the bottleneck is shifting from writing the code to taking code to the production faster. The organizations who did four releases in a year, now they can do 40 releases in a year for that. Second big point really which really comes is the role of developer is changing. Previously developer was more focused on writing the code.
Now they are more focused validating the code and taking code to the production faster and in a much scalable way into the deployment. Second big impact which AI is really bringing is the agentic future. Today, maybe 30 to 35 million developers who are going to use the platform or will use the platform, but the agents
have already outnumbered the developers or going to outnumber, it’s pretty likely next year there will be 100 million agents will be using the platform. They will require the same infrastructure. They will require API. They will governance for that. So that’s a huge part of that. Next big AI is really driving is the last six months you saw the conversation changing from token maxing to the token economics for that. It is top of the mind.
The last one which AI actually drove as driving platform engineering lot is the sovereignty and the compliance for that, putting the guardrails over here. So AI has an implication changing the developer personal behavior, changing the frequency of the code development, bringing the new developers
It has an impact on FinOps. It has impacts on the security for that. And the bigger impact is that AI gives a brand new persona which platform has to cater with the agents for that. So this is the perfect storm. And this perfect storm cannot be at more appropriate time where a lot of other things are happening. The sovereignty requirements are growing for that. You have seen the hardware pricing is increasing so much. So you see how AI is creating the
DRAM pricing challenges and the server costs for that. now the AI is also driving FinOps. Previously FinOps was bolt on, now it has to be integrated. So wherever you see that AI has impact which is just avalanche across the whole platform engineering.
Stephen O’Grady (12:33)
so you know you know, Pankaj, you you mentioned that the you know, the sort of ballpark hundred million number could be less, could be more. you know, who who knows? It’s a ballpark. The the net of it is we’re gonna have a lot of these agents running around interacting with these systems. So how do you think about this as a persona, right? We’re we’re used to personas like de developer, operator, admin, etc. This is a new persona, you know, from an agent standpoint. So like where does the platform fit in here?
Pankaj Gupta (13:01)
So agents will be using the platform for writing the code, executing the code independently. So that’s the biggest, biggest impact. And their frequency of writing the code and executing the code may be exponentially higher what the human can do that. That’s the first part. Second part, they will have the unique needs for that. They will require the APIs, they will need the models, they will require
More importantly, what they will require is the traceability. They will require very strong guardrails for that.
You heard a few months back, it brought in one of the company’s infrastructure down for that. You saw an example that one of the agent deleted their customer data for that. the requirement for guardrails is going to be so significant over here for that. Observability is going to be very huge. The audit is going to be pretty huge for that. Because there is no human interaction. The agents can do whatever at 3 AM for
that and then you have to figure out what happened for that. So that’s the reason there is so much brand new conversation in the industry for last 8 to 12 months about the MCP server, the MCP definitions and their requirements and the execution is just…
changing or evolving to bring lot of guardrails into that. Same is conversation associated with the AI gateway. So now, platform engineering’s biggest job is to understand that what is the roadmap for their agents look like? What are their unique needs for that?
What is the guardrails they have to put it? How they are going to troubleshoot it? How they are going to put the traceability or audit? Audit is going to be huge, huge for that. And all these things has to be done at a scale. They’re not just 30 developers or 100 developers who are using. This is the agents are doing, which is autonomous for that. So I think this is a very big persona. And this is the first time platform gets a non-human persona.
Stephen O’Grady (15:22)
Okay. N you know, when we talk about that, how do we engineer for that? And and you and I have have sort of talked about this, you know, sort of as as five pillars. So,
Pankaj Gupta (15:33)
Yes.
Stephen O’Grady (15:33)
you know, we don’t have time to go through them at any level of of great detail, but can you give us sort of a quick sort of recap of of the
Pankaj Gupta (15:39)
Sure.
Stephen O’Grady (15:40)
you know what you consider the five pillars to be?
Pankaj Gupta (15:42)
The first one is the AI native platform. So the platform has to support AI workloads and agents natively into the platform. It could be the GPU provisioning, self-service for GPU, it could be the MCP server, it could be the AI server, the virtualization list goes on and on for that. Second one is the platform has to evolve
from a single developer focused persona to multi-persona experience for that, which includes the agents, of course, as non-human. It will include the business leaders who are going to look at ROI, the FinOps cost predictability, the security and the compliance team who are going to move from a static compliance one in a year to…
every day or every hour compliance or every action compliance. Third one is the platform engineering has to evolve from a bolt on a FinOps to more embedded FinOps, so putting the gates when you provision the environment for that and constant monitoring, not just the regular workloads, but also the AI workload and a stronger security to compliment shift left.
security shifts down into the platform, security comes into the runtime for that, and how AI is expanding the security gap, that has to be addressed for that. That’s a brand new attack surface for that. And the last one is that composable by design. The platform has to evolve to support new components, and more and more choices is going to come into here. So you should be able to replace one CI/CD tool with another CI/CD tool, one model with another model.
one gateway with another gateway without effects going through the rest of your stack for that. It is a tall order, but those are the five things together frame the plate from engineering.
Stephen O’Grady (17:47)
Okay.
So that all makes sense. as we wrap up though, you know, I would like to I’d like you to leave the audience with sort of your ideas. Like if if I’m a platform engineer, sort of what are the next steps? Like what should I be looking at? What should I be thinking about?
Pankaj Gupta (18:01)
I think there are two things. First and foremost is familiarize yourself where the platform engineering evolution is happening. So Broadcom has published a Platform Engineering 2.0, a white paper which is vendor neutral, talks about the industry trend, what the platform engineering has to evolve. That’s one is published on our website. Just search Platform Engineering 2.0.
Second action is that go back to your organization and look at these five pillars, which has the more urgency for that. Most probably, every organization is looking at how they enable the AI workloads on this platform. Look at is it the AI has to be supported natively, if they have already done it, how they are addressing multi-persona, or if they are behind on FinOps. Figure out where you are starting point for that.
but start looking at platform in a much broader perspective into your organization for that. Also bring infrastructure teams with you, the people who manage the servers, the compute, the networking, the storage with that, because they remain the foundation for every platform. And keep learning because this field is going to evolve for next two or three years very significantly.
Platform engineers have the right foundation and already know many of the things with platform engineering will continue to deliver and continue required.
Stephen O’Grady (19:38)
Can’t close on a better note than that. because particularly in this day and age, as quickly as things are moving, you have to keep learning. so with that we’ll we’ll close out. I really appreciate everybody’s time and Pankaj, thank you so much for joining us.
Pankaj Gupta (19:50)
Thank you Steve for having me.
