What does it take to put AI agents to work? AWS CEO @mattsgarman sat down with @AlexeHeath on the Sources podcast to share what he's seeing across millions of customers, from the largest banks to two-person startups.
First, agents need data. To do real work, an agent has to reach all of a company's data, and for many companies some of it still sits in a mainframe or a legacy data center. So customers are speeding up cloud migrations they had once put off. Matt calls it a flywheel: the more core workloads move to the cloud, the more companies can do with AI.
Agents also need infrastructure, and AWS builds it to be a good neighbor. As Matt put it, "all data center producers are not created equal." AWS works with local and state leaders, brings high-paying jobs, and pays for the power it brings to the grid.
And agents need to be trusted. AWS is investing in automated reasoning, which uses mathematical proofs to show an agent will follow the rules it's given. Matt's example: an insurer could let an agent process claims on its own, so customers get an answer in 10 seconds instead of waiting a month.
Listen to the full conversation: go.aws/3VkrwOr
Nov 11, 2026 · 5:12 PM UTC
