Veera Ravindra Divi

From Autonomy to Assurance: Governing Agentic AI for Production Enterprise Systems

Abstract:

We are entering an era in which software no longer just follows instructions — it decides, plans, and acts on its own. Autonomous AI agents are moving out of demos and into the systems that run real businesses, and with that shift the central question changes. For a decade the race in applied AI has been about capability: can the model do it? The question that now determines whether these systems earn a place in the real world is assurance: can we trust what they do when no one is watching?

 This keynote charts the journey from autonomy to assurance as the defining challenge of the next phase of applied intelligence. It argues that trustworthy autonomy is, above all, an engineering and architecture problem — not merely a smarter-model problem — and that the discipline we build around these agents will matter as much as the intelligence inside them. Drawing on lessons from deploying AI in mission-critical, high-stakes enterprise environments, the talk offers a practical way of thinking about how organizations can grant machines real autonomy while keeping security, accountability, and human oversight intact.

Key Takeaways
- Treat every autonomous agent as an adversarial optimizer; design for the path it can take, not the one you hope it takes.
- Separate the reasoning plane (proposes) from the control plane (disposes); govern reachable blast radius, not individual tool calls.
- Grant autonomy in stages earned by evidence, not by default.
- Use multi-agent debate and LLM-as-judge validation as governance and verification oracles, not just generators.
- In mission-critical enterprise systems, assurance — security, auditability, reliability — is the real adoption blocker, and it is an architecture problem.

Profile:

Veera Ravindra Divi is a principal-level software engineer, independent researcher, and IEEE-published author working at the intersection of agentic AI, distributed systems, and mission-critical B2B commerce. He brings more than 11 years of experience building low-latency trading systems at Wells Fargo and designing scalable distributed systems, AI orchestration platforms, and B2B marketplace architectures at Amazon. His research spans autonomous LLM orchestration, multi-agent governance, trustworthy generative AI, and human-assured agentic interoperability for resilient enterprise systems. He serves as a technical program committee member, reviewer, and session chair for international IEEE conferences, and is an invited speaker at venues including the 2026 IEEE World AI IoT Congress. He holds an M.S. from a U.S. university.