The Future of Autonomous Systems with AWS VP Swami Sivasubramanian

The Future of Autonomous Systems with AWS VP Swami Sivasubramanian

Oct 21 · 28 min

People

  • Tim Porter
    Tim Porter Host
  • Swami Sivasubramanian Guest

Watch on

Suggest an edit

​@amazonwebservices VP Swami Sivasubramanian explains why infrastructure, context, and correctness decide which enterprise AI agents reach production. Recorded live at the 2026 IA40 Summit, he tells Madrona Managing Director Tim Porter how Amazon teams building agents kept hitting the same wall until he told them to stop and build the infrastructure once. In this conversation: + Why every Amazon agent team needed micro VM isolation, identity, and observability before rollout + The thesis behind DynamoDB, SageMaker, and Bedrock: remove undifferentiated heavy lifting so builders can focus on what only they can do + Why he expects context as a service to become a primitive for the next twenty years + How neuro-symbolic guardrails can catch rule violations without suppressing an LLM’s creativity + Why value will be captured at every layer of the stack + Why Swami calls his approach to agent reliability “trust in depth” + Why robotics is about to enter a golden age and what physical AI still lacks For founders and engineering leaders building agents for enterprise customers, this conversation lays out the hurdles between a working demo and a production deployment, and which of them a platform can remove for you. Chapters: 00:00 – AWS VP Swami Sivasubramanian on Leading Agentic AI and Emerging Technologies 02:04 – How a PhD Internship Project Became Amazon's Dynamo Database 03:09 – The Undifferentiated Heavy Lifting Thesis Behind DynamoDB, SageMaker, and Bedrock 05:52 – Why Only 7% of Enterprises Have Deployed AI Agents 07:44 – Why Amazon Teams Stopped Building Their Own Agent Infrastructure and AgentCore Began 10:02 – Why Value Will Be Captured at Every Layer of the AI Stack 13:27 – Why Context as a Service Could Be a Primitive for the Next 20 Years 14:14 – How Neuro-Symbolic Guardrails Keep AI Agents Inside Enterprise Rules 17:13 – How Swami Built a Paxos Service for Dynamo in Under Three Months Using TLA+ 20:43 – Why Robotics Is About to Enter a Golden Age and What Physical AI Still Lacks 23:53 – Why AWS Open Sourced Strands 26:23 – How Specialty Inference Providers Fit Alongside Bedrock

AWS Grok IA40 Summit Madrona

Chapters

  1. 0:00 AWS VP Swami Sivasubramanian on Leading Agentic AI and Emerging Technologies
  2. 2:04 How a PhD Internship Project Became Amazon's Dynamo Database
  3. 3:09 The Undifferentiated Heavy Lifting Thesis Behind DynamoDB, SageMaker, and...
  4. 5:52 Why Only 7% of Enterprises Have Deployed AI Agents
  5. 7:44 Why Amazon Teams Stopped Building Their Own Agent Infrastructure and...
  6. 10:02 Why Value Will Be Captured at Every Layer of the AI Stack
  7. 13:27 Why Context as a Service Could Be a Primitive for the Next 20 Years
  8. 14:14 How Neuro-Symbolic Guardrails Keep AI Agents Inside Enterprise Rules
  9. 17:13 How Swami Built a Paxos Service for Dynamo in Under Three Months Using TLA+
  10. 20:43 Why Robotics Is About to Enter a Golden Age and What Physical AI Still Lacks
  11. 23:53 Why AWS Open Sourced Strands
  12. 26:23 How Specialty Inference Providers Fit Alongside Bedrock