Not Just More Pilots: Shweta Gummidipudi on Scaling AI at Twilio
Technovation with Peter High (CIO, CTO, CDO, CXO Interviews)Show Notes
Most companies can prove an AI use case works. Scaling it is where they stall. Shweta Gummidipudi, CIO at Twilio, has spent her first year building the infrastructure layer that makes scale possible: a single model gateway for routing and cost control, a shared library of agents and connectors, and inference budgets that keep business-unit builds inside enterprise guardrails. She also addresses two problems most governance conversations skip: agentic identity, which the industry has not solved, and agent sprawl, which she maps directly to the shadow IT problem CIOs spent the last decade managing. Alongside those structural challenges, she describes Twilio's Customer Zero program, where IT runs Twilio's own products in live workflows and converts that experience into reference architectures that enterprise customers adopt faster than vendor documentation alone.
Key Highlights:
Why a single model gateway is the prerequisite for moving from pilots to scale
How Twilio structured a three-tier access model to prevent agent sprawl without blocking experimentation
Why agent identity is categorically different from human identity and what that means for security
How the Customer Zero program turns internal practitioner experience into customer reference architectures
Why career-leveling guides need to be rebuilt from scratch rather than retitled for the AI era
More from Technovation with Peter High (CIO, CTO, CDO, CXO Interviews)
AI efficiency is not the destination. At Sunlife, it is the means to something harder to automate: meaningful human connection.
AI layered onto a broken process doesn’t fix the process; it just runs it faster. Prakash Kota spent his first 18 months at UKG consolidating two legacy companies into unified CRM, ERP, and data systems before any AI conversation began. Now, as Chief Transformation Officer, he is shifting the question from “what tools can we deploy?” to “what outcome are we trying to reach?” That discipline extends to the customer funnel, where AI-assembled signals travel with a prospect through every handoff, and to sales reps, where an autonomous assistant pushes pre-meeting briefs before a rep has to ask, reclaiming the 40–50% of call-prep time currently spent on research.
The binding constraint on AI adoption is not access to tools. It is the organizational fear and mindset shift required to reimagine work entirely.
Healthcare has accumulated friction that patients have come to accept as normal. Novant Health’s Chief AI Officer is systematically dismantling it.
Most generative AI programs stall at the pilot stage. Zoetis reached 95% computer-based colleague adoption.