We let our cloud agents query prod (with OIDC, not secrets)

Formal’s approach gives cloud coding agents access to production and development systems through OIDC rather than reusable, long-lived secrets. An agent obtains a short-lived identity token from its platform; Formal verifies its signature and configured claims, then applies policies that can use details such as the repository, agent session, and initiating engineer. Integrations can also issue short-lived Connector tokens, and support SPIFFE identities. A machine identity is cryptographically verified, while any human identity attached to it is trusted only because the customer has chosen to trust the platform asserting that claim. This distinction lets teams treat an agent run as a temporary, attributable delegation, not a permanent machine account.

Formal uses this model to let agents query production analytics databases after requesting human approval through Slack; access is read-only, short-lived, and limited to the approved session and database. Agents can similarly investigate asynchronous workflow runs using approved, logged access, with requests attributed to both the agent and its owner. For development, they can federate into Formal and cloud-provider test environments to deploy and verify changes against real infrastructure rather than mocks. The central argument is that agents become more useful when they can gather data, diagnose failures, and validate work themselves, but this capability need not mean handing them static credentials. Production access stays constrained and approval-gated, while development permissions can be broader without becoming permanent.