We’re building Swytchcode: an execution layer between AI agents and enterprise APIs

We’ve been working on a problem that keeps coming up as AI agents move into production: What happens between an agent deciding to do something and the API actually executing it? The obvious architecture is: Agent → Salesforce / SAP / Oracle → Done But production adds a lot more: Authentication Permissions Validation Retries Idempotency Rate limits Auditability Putting all of this inside every agent gets messy pretty quickly. So we’re building Swytchcode as an execution layer: Agent → Swytchcode → Enterprise API The agent handles reasoning and intent. Swytchcode handles the execution side and the controls around it. We’re particularly interested in APIs where one agent action can trigger a lot of downstream side effects. How are you handling this today? Are your agents calling production APIs directly, or do you have an execution/policy layer in between?