KM4BI KM4BI

Cloudflare, Stripe, and Okta decide whether your agent ships

Production AI agents increasingly depend on runtime, identity, data, payments, observability, and kill-switch infrastructure.

Nate B. Jones argues that models are only one part of the agent economy. The companies that increasingly determine whether an AI agent can ship are the ones controlling the environment around the model: Cloudflare for runtime, Okta and Oso for authority, Snowflake and Databricks for governed data, Stripe for payments, and Datadog or LangSmith for observability.

Control infrastructure becomes the strategic layer

A production agent needs state, memory, tool access, recovery, approvals, spending limits, and auditability. That pushes value toward infrastructure that can answer seven operational questions: where the agent runs, who it acts for, what it can know, what it can change, what it can spend, what gets observed, and who can stop it.

Infrastructure vendors define the boundaries

Cloudflare, AWS, and Vercel are turning runtime into a control surface. Okta, Oso, WorkOS, Entra, and AWS Agent Core Identity are converging on delegated authority. Snowflake, Databricks, and BigQuery/Gemini want agents to reason inside governed data environments. Stripe and the card networks are extending institutional trust into agentic transactions.

Observability and kill switches become product features

Traditional logs are not enough to know whether an agent respected user intent, selected the right tool, used the right data, or created an expensive loop. Teams need traces, costs, tool calls, retrievals, and evals stitched together. Stopping an agent also requires more than telling the model to stop: runtime, identity, gateways, payments, and workflow layers all need enforcement points.

What teams should do next

Pick one concrete workflow — support, refunds, claims, usage review — and map the control surfaces before production. Any unanswered row is a production risk, especially when agents can work around permission systems originally designed for humans.

Source