Not Just More Pilots: Shweta Gummidipudi on Scaling AI at Twilio

0:00Everybody gets excited with AI, but at the same time, we want to do it scalably. We want to do it with safe velocity, and we want to do it responsibly.

0:11Peter High: Welcome to Technovation. I'm Peter High, president of Metastrategy. My guest today is Shweta Gummidipudi. Shweta is the chief information officer of Twilio, a customer engagement platform connecting businesses with their customers through messaging, voice, email, and data-driven experiences. Twilio generated approximately$5.1 billion in revenue in 2025. Shweta leads global information technology strategy, enterprise engineering, and AI and data-driven innovation. She joined Twilio in October of 2025 after leadership roles at Snowflake and NVIDIA, bringing more than two decades of technology experience to her current post.

0:47Peter High: In our conversation today, we'll cover her first year at Twilio, the transformation of its opportunity-to-revenue process, and how she's building the data foundations needed to scale artificial intelligence. We'll also discuss Twilio's Customer Zero program, which tests its own products internally, and her approach to managing agent sprawl, evaluating AI investments, and demonstrating business value, among other topics we'll cover. Shweta, welcome to Technovation. It's great to speak with you today. Great to be here, Peter. Thank you for having me. And now for a word from our partner, Blitzy. Want to shift software five times faster?

1:19Peter High: meet Blitzy, the autonomous software development platform built for enterprise-scale initiatives. Your code base has outgrown any one team. Tens of millions of lines, decades of technical debt, more than any other AI tool can hold in its context window. Blitzy starts by understanding. Thousands of specialized agents map your architecture, dependencies, and business logic across every language into a knowledge graph. That graph also services hidden CVEs and zero days before an alert ever reaches your desk. Then you describe what you need in plain English and approve the agent action plan Blitzy builds.

1:55Peter High: Grounded in the knowledge graph, Blitzy executes entire engineering epics over weeks of continuous compute and every line is tested end-to-end before it reaches your team. Modernize legacy systems, close critical vulnerabilities, ship new features five times faster. Learn more at blitzy.com slash technovation. That's B-L-I-T-Z-Y dot com slash technovation. And now on to the interview. That's a pleasure. Let's begin with Twilio's business, if you would, Shweta. I think a number of people who'd be watching or listening to this would have some familiarity with it. But for those who may be less familiar, I would love to have you describe the business you helped lead.

2:35Absolutely. I think it's a privilege and a pleasure to be representing Twilio today. And I can say that most people use Twilio every day without even knowing it. So when DoorDash texts you that your driver is in the way or when your bank sends you a login code or when an airline calls about a gate change, there's a good chance that it's us behind the scenes. We are the infrastructure behind customer communications for hundreds and thousands of brands. what we have done for the last 17 years is to take something messy like the global telecom with its carriers regulations and many different channels and turn it into a few lines of code that a developer can use so messaging voice email verification and the customer data that makes all of those conversations relevant that is our core business what's changing now is who's in the conversation.

3:36It used to be a brand talking to a person. Now it's people talking to AI agents and agents talking to other agents. Those interactions need to remember context. They need to be personal. They need to be trusted. So we are building that layer, secure, reliable infrastructure with the data that gives every conversation like memory and intelligence. And my job as a CIO here is to make sure that Twilio runs on that future first. I know you alluded to customer zero, so we do that a lot with our products.

4:12Peter High: Amazing. What a great overview. And a number of things you mentioned there that I'm looking forward to double-clicking on. Before we get there, though, I'd love to have you describe for our audience your purview as chief information officer. What's within your realm of responsibilities? Yeah, happy to. So as you'd mentioned, I've been with Twilio for a year now. And as the CIO, I own everything that happens from a technology and automation standpoint inside of the company. And it's all around how our people work and how Twilio adopts AI internally. So the biggest priority, obviously, for all CIOs and CXOs these days is AI.

4:56And with respect to AI, I have three big priorities. One is the shared AI infrastructure so teams can experiment without being disconnected in a sea of tooling. The second is all about governance that keeps up with the speed of AI. It's a very important initiative for us internally. And last but not the least is all about being customer zero, which means that we run Twilio's own products and our own real internal workflows. and we test them, we harden them even before our customers get hold of them.

5:35Peter High: Very interesting. You mentioned shared AI infrastructure for experimentation. Talk a bit about the AI infrastructure that you have helped assemble. I'd be interested in how you've thought about the combination of what the organization builds versus what it curates. How would you define that for us? Yeah, I think it's very, very pertinent and sits at the heart of the AI journey, you know, proving that one use case works with AI is an easy thing. Scaling is the hard part. How do you move from pilots to scale? And infrastructure and getting the infrastructure for AI right is very key and essential to that.

6:18I usually end up asking a few practical questions. It's things like when you assess infrastructure readiness. It's things like, can you see what your agents are doing? Can you control costs? Can you send each workload to the right model? Now, the answer to any one of them is no, you aren't ready to scale, right? You're just in the modality where you're running more pilots. So when we think about building shared infrastructure, we are talking about how every model call goes through one gateway, which gives us routing, cost controls, and all the privacy controls that we need. Tools and data are reached through one standard auditable path, and teams share a library of agents, of skills, of connectors.

7:07So a second team reuses what the first team already built rather than rebuilding it. That's a very critical part of creating scalable infrastructure. And our approach, Peter, has been on governed federated build models. So our enterprise technology owns the platform, the trust standards, the security, the privacy, the cost tracking, the telemetry that's associated. And business leaders still have the flexibility to do a little bit of the build of their own use cases within inference budgets.

7:43Peter High: which takes us a bit into the second area of responsibility that you mentioned governance that keeps up with the speed of ai and what an interesting framing of that given the pace of change that's happening right now given no doubt the tremendous appetite that a company like yours has for the use of it also can you tell us a bit more about some of the governing philosophy that you have for governance itself to ensure that it is in fact keeping up with the speed and pace of change relative to AI? Yeah, that's a fantastic question. You know, how do you govern AI without slowing everything down, right?

8:18We found initially that risk review, not technology, was what slowed us down. So we are redesigning a third-party risk management for AI. We made a gate for proof of concepts, for instance, to be lighter than the gate for production. The second part is also guardrails. We have put controls into the platform itself through a single auditable path to all of our tools and data. And we are looking at more robust runtime as well as agentic identity, because agentic identity is very important for us to be able to actually do more use cases, especially around things like writing. Asian identity is a separate thing from human identity.

9:07The industry itself hasn't fully solved yet. And until we have really good answers on identity and runtime controls, we will need to keep Asian building contained. So one of our approaches has been how can we seed this with the technology that exists today so that we are not actually slowing things down while keeping our build contained.

9:35Peter High: Very interesting. I do want to mention you and I both have now mentioned your Customer Zero program. Talk a bit about that, the philosophy behind that. I can certainly understand the power, the extent to which one of the personas that Twilio serves are, in fact, your peers in other organizations. So the extent to which you can replicate that experience prior to them, you can offer invaluable insight into what's working, what isn't, what additional functionality might be there. I hope I'm not grossly oversimplifying the nature of a Customer Zero program, but I'd love to have you share a bit about the power of that in a role like your own.

10:10Yeah, I mean, it's a great question and something that's super close to my heart. I feel as a CIO, most CIOs today have to lead as Customer Zero for their company's own transformation. Our jobs have moved from running systems to redesigning how work gets done. And a very integral part of that is owning the platform, the guardrails, and being honest about what your own technology can do and cannot do and feeding that back to the product roadmap. customer zero at Twilio we are builders we have a very deep culture of building we speak about building all the time but we don't just talk about it and customer zero is where our you know that culture actually meets reality I think of it as a couple of pillars the first pillar is all about experimentation on the platform building on the platform giving feedback and hardening carding it.

11:13The second pillar is all about actually creating value through our platform and measuring that value, getting, you know, it's super quantified and qualified so we can actually chase value. And the third piece is, you know, and this is the cool part. This is where we enable our customers with reference architectures, with blueprints, so that they can accelerate their own journeys with more confidence. And as a practitioner myself, when I, you know, talk about a reference architecture, I talk about a blueprint, I think there is an inherent credibility in that equation. And you see, you know, the pure CXOs take to that much faster.

12:05So it's all about enabling our product to become the best that it can be, unlocking the most value out of our product and helping our customers actually adopt our product faster and unlock value from it.

12:22Peter High: I like the inclusive nature of your answer at the outset as well, that this is an opportunity for many CIOs to be the first experimenter with so much innovation that's coming about and drawing conclusions on behalf of the rest of the organization before perhaps advising the rest of the company on where to run towards as opportunities present themselves, for example. A phenomenal set of perspectives there. Thank you for that. I want to return to a Fascinating point you made earlier, Shweta, which is that so much business historically, including in an organization like Twilio, has been customer to business, customer, human to human.

13:00Peter High: And you talked about the evolution to customer to agent and then increasingly agent to agent. And this is a profound change, certainly. And I wonder, it's still early days for anyone in an organization like yours as well, but I wonder what considerations you would advise all of us as customers, but certainly for people who work for businesses that sell to others where the agents considerations are going to be increasingly important. How do you think about what analogies there are from traditional ways of doing things that hold versus those aspects that will be fundamentally new in an environment where we have human to agent and agent to agent interactions like you described?

13:44Yeah, that's a very topical, very important question that I guess every company is grappling with. Maybe one way to answer that is to look at some specific use cases. So we actually talk about what is needed, what is changing, what is needed, and how we get the value from our technology to handle this new way of being and new way of, you know, communicating. So if you take something like, you know, signups in a campaign, every company launches a campaign and there is, you know, signups associated with a campaign. One of the things that everybody is seeing with increased agent activity is there could be fraudulent signups and there could be a lot of fraudulent agentic signups, which are not human in nature.

14:40That could be a use case. Another use case could be that, you know, you are in the middle of a bank transaction in a highly regulated industry. And suddenly there is a possibility that you could get hijacked by an agent and, you know, it could result in a financial transaction being compromised. How do you think about scenarios like this? And how can you think about defending yourself around scenarios like this? And the world of regulatory compliance, for instance, requires like immutability. Every transaction that you perform has to have like an audit trace associated with it and very definite identification of who is actually performing these transactions.

15:28So the things that we have implemented at Twilio, for instance, is we have used our own identity suite of products. and we are looking at how we can prevent these types of fraudulent transactions. Again, goes back to everything that I said about internal AI as well. I think it's the guardrails around identifying what is human versus agentic and when it is agentic, applying the lens of heightened scrutiny and security around it and making sure that you filter it, you understand the behavior, you're able to isolate the signal from the noise. It's just that new sensibility that every organization has to develop and make more robust.

16:24Peter High: Shweta, you've also written about the notion of agent sprawl as a new form of shadow IT. I wonder how do you, thinking about past considerations of overcoming the notion of shadow IT, again, how closely the analogy holds from your perspective, and how do you give teams room to experiment while maintaining visibility, consistent controls, clear ownership, these sorts of things? Yeah, that's a hard one because everybody gets excited with AI and everybody wants to build and we are a build culture and we don't want to say no. But at the same time, we want to do it scalably. We want to do it with safe velocity and we want to do it responsibly.

17:08So we the way we think about it is, you know, three tiers. Everybody gets better search and everybody gets a personal AI assistant. and there's like a cost associated with that that we cap. Teams with repeat work get agentic workflows built for them. So that's where shared skills, an AI marketplace, an AI studio where you can download the skills and the workflows that you repeatedly need or invoking the agentic workflows from the place where you work today becomes very important. Then we have a center of excellence, a group of builders, which is a cohort of people within our business verticals who build more extensively for these shared, shareable team assets within guardrails.

18:01So that's deliberate. While we want everybody to do their own AI, at the same time, we don't want the same AI solution, different flavors of it to be built by a thousand people within the company. So that's how we're thinking about Sprawl and how to keep it contained.

18:20Peter High: Yeah, very interesting. You're also sponsoring a major transformation of Twilio's opportunity to revenue process. And I wonder if you could describe the process and where you found the greatest friction while redesigning the journey to better serve customers and ultimately accelerate business. Yeah, that's been an incredible experience. It's a pretty complex journey that even predates, honestly, me joining in. but I now have the opportunity to helm a part of that. This was all about, so Twilio grew very organically, acquiring different businesses over time. And when companies do that, there is often technical debt associated with having multiple platforms in your ecosystem.

19:11And when you're innovating at the pace that Twilio is innovating, especially at the agentic era, there's always a pressure to bring new features to market and have the ability to serve your customers through common, you know, invoices, a standard way of invoicing and a standard way of exposing your feature sets. and to enable that we undertook a journey, a multi-year journey to simplify our lead to revenue process. A big portion of that is also redesigning how we do business and simplifying that. We have applied automation, we have applied AI and we are basically reimagining how the entire process flow works.

20:04And, you know, I'm glad to report that we already have a few customers live on the new way of doing things, and we will continue to expand on that. Yeah, but it's been an interesting journey. These things are always, you know, complex and so mission critical.

20:22Peter High: I wonder also in light of so much of the change we've already discussed, Shweta, how have you thought about the skill mix necessary to anticipate the changes that have arisen, but they will continue to come? How do you see the evolution of your team and the skills that are on the rise as a result of some of the change you described? Yeah, you know, Peter, the tools are the easy part. It's the skills and the people and bringing the people along. That's hard. I'm sure you appreciate that. One of the things that we are trying to do is companies traditionally have career leveling guides, which describe how different functions, you know, progress and the mix of functions within a certain vertical.

21:10I have, for instance, an IT career leveling guide. I think it needs to be reimagined in the context of AI. What does that look like? What does an AI guide look like? What does the AI job family mix look like? And, you know, that's how we are approaching it. We want to be the gold standard for an AI-enabled, AI-native organization. With that view, we have started mapping how our processes as well as our roles evolve. In the past, if we had like a business system analyst who was working on, say, a go-to-market workflow, go-to-market automation workflow, the future state for something like that would be like, you know, they would have an AI assistant working for them, and they would be a driver or a leader of that assistant that helps them actually do portion of the requirements and the design.

22:14So their role changes from being just, you know, an analyst to like an operator of AI, who has a lot of analytical abilities and agents working for them. We're trying to do that for every function, not just my organization, but the rest of the company as well.

22:36Peter High: And how do you think about change management? The outside of your response, such a critical point you raise that so much of what we're describing is a human set of changes that need to take place in order for this to fully blossom. How do you think about the concepts associated with change management as a catalyst for the change to come? Yeah, I think companies sometimes underestimate the structure that is needed to do proper change management in the face of AI because they assume that the benefits will be apparent and people will like see it and use it and adopt to it and adapt to it. I actually have an enterprise enablement function with a leader dedicated to that, where we do it in different phases.

23:19So the first phase, depending on maturity of your company, is usually like getting people literate. They may not understand all the various avenues in which you can consume AI. So literacy is like the very first step. And then how do you start from that beginning of literacy and transition people to fluency, where they're using tools with great ease? they're actually thinking about redesigning the workflows that's that's a journey and another thing that I would call out as we are going through this enablement exercise is that even within the same company one size doesn't fit all you will typically see that your marketing teams or your sales teams are already entrenched in a lot of this and are willing to go with higher velocity than some of the back office functions.

24:15So there's always that element. The readiness itself, even if we had the tooling, the data readiness, the infrastructure readiness might be variable. So these are all the factors that you have to consider when you're thinking about enablement. Yeah.

24:32Peter High: Wise advice, certainly, Shweta. I wonder also, as you think about the necessity to remain current on the change that is happening, changes to come. Is there anything you read or listen to that has been particularly influential for you that you would recommend to others? How do you stay current, if I may? Oh, wow. That's a loaded question because I feel like there's not enough hours in the day to stay part of the AI. I wake up every morning and I think it's true for most of us, we learn something new because the pace of change is so dramatic and it is so, you know, constant. I do consume, you know, a lot of sources of information, whether it is through LinkedIn or podcasts or news agencies.

25:28But for me, I think what has been most valuable is to get my hands dirty and to actually experiment and learn. I feel after a very long time in my career, I feel truly, truly very excited about technology. Again, I think it's a rebirth, reawakening for a lot of us in tech where the possibilities are so limitless. So I'm always tinkering. There are things that are sanctioned within the environment that I'm responsible for. But I also have a parallel environment that is my personal environment. And I am constantly like downloading things and playing with things. There's example is like Muse and Instinct, And even though we haven't sanctioned them yet within our enterprise, I love to play with them in my sandbox and learn through that.

26:28I think that for me is the most informative avenue and a way also to have a very practical perspective on where this industry is headed.

26:40Peter High: That makes a lot of sense. Well, Shweta, thank you so much for a fascinating conversation. It's very interesting to learn more about a consequential first year with Twilio and some of the things that you and team are working on to help define the future of the company. Thank you so much for a fantastic conversation. Thank you, Peter. I really enjoyed speaking with you. And I appreciate the opportunity to, you know, represent both Twilio and the incredible work that my team is doing with Twilio. Thanks for listening to Technovation. If you enjoyed this conversation, subscribe and leave a rating.

27:13Peter High: It helps more leaders find our podcast. Thank you and I look forward to seeing you next time.