Descriptions:
Nate B Jones argues that the real power brokers in the AI agent economy are not model companies like OpenAI or Anthropic, but a layer of infrastructure providers that determine whether any agent actually reaches production. The video identifies five control-point categories and the companies dominating each: runtime (Cloudflare’s Agents SDK with durable objects and built-in SQL, AWS Bedrock Agent Core, Vercel’s AI gateway), identity (Okta’s recently launched Okta for AI Agents, Auth0, WorkOS, Microsoft Entra), data governance (Snowflake and Databricks with their governed semantic layers), payments (Stripe and its Agent Commerce Suite, which spans fraud, issuing, billing, and treasury), and observability (Datadog’s LLM observability platform).
The core thesis is that a stateless model — one that returns a response and forgets the conversation — cannot run a production agent. Production agents require durable state, scheduled execution, tool orchestration, identity delegation across third-party APIs, spending authorization, and real-time observability. Each of those requirements hands control to an infrastructure layer that sits beneath the model and above the end user.
Jones draws on specific recent announcements: Cloudflare’s Agents Week, Okta’s April launch of agent identity features, Stripe’s aggressive expansion into agentic commerce, and Databricks’ Mosaic AI agent framework. For engineering and product teams deciding where to build, the video functions as a map of the emerging agent infrastructure stack and the vendor lock-in dynamics that come with each layer.
📺 Source: AI News & Strategy Daily | Nate B Jones · Published May 20, 2026
🏷️ Format: Deep Dive
