Seven questions decide whether your AI agent ships. Most teams can answer two.

Source: Nate’s Newsletter Date: 2026-05-20 URL: https://natesnewsletter.substack.com/p/agent-infrastructure-control-layer

Summary

Nate’s Newsletter argues that model capability is no longer the binding constraint on agent deployment — infrastructure is. The piece frames six control-layer questions that production agents must answer: where the agent runs, what state it retains, whose identity it acts under, when it needs human approval, what it can spend, and who can stop it. The thesis is that Cloudflare, Stripe, Okta/Auth0, and Datadog are the actual gatekeepers deciding whether agents reach production, not model vendors. “The boring layer is where the power is moving.”

Implications

• Agentic engineering: The control-layer framing is a useful taxonomy for agent readiness audits. The six questions map directly to infrastructure categories (runtime, memory/state, IAM, approval workflows, payment rails, observability) that are currently handled by different vendors with no standard integration layer — a gap the piece identifies as the primary blocker.
• AI ecosystem / power dynamics: Cloudflare, Stripe, Okta, and Datadog being named as the real gatekeepers is a significant power-dynamics observation. If those companies develop agent-specific primitives (agent identity, agent spending limits, agent audit trails), they accrue leverage over the agent economy regardless of which model is inside. Watch for infrastructure companies shipping “agent-aware” tiers.
• Agentic engineering: The claim that most teams can answer only two of the six questions is a useful benchmark for where the industry actually is on agent maturity — well past prototype, still far from reliable production deployment at scale.