Your infrastructure doesn’t stop at deployment.
Neither does the intelligence watching it.

Day 0 is planning and design. Day 1 is initial deployment. Day 2 is everything that happens after: keeping environments running correctly, managing configuration changes, controlling costs, maintaining compliance, and ensuring environments stay aligned with what was intended when they were deployed. Day 2 is where most infrastructure problems occur and where most infrastructure cost accumulates. It is also where most organizations have the weakest tooling, relying on manual processes, scheduled scans, and reactive incident response. Operate replaces that with continuous, automated, real-time governance.

Operate continuously compares the live state of every running environment against the IaC specification stored in Git and indexed by Curate. When a configuration item changes outside the governed process, whether through a direct cloud console action, a manual modification, or an out-of-band script, the deviation is detected in real time and surfaced in the operations dashboard immediately. The detection is not based on scheduled scans. It is continuous. The dashboard reflects drift the moment it occurs, with specific detail on what changed, what the expected value is, and what depends on the affected resource.

Both options are configurable. Platform teams can configure automatic remediation for known, low-risk drift patterns, where Operate detects the deviation and immediately restores the environment to its approved configuration without human intervention. For higher-risk changes, or environments where human sign-off is required by policy, drift triggers an approval workflow that routes to a designated approver. The approver can accept the change (updating the IaC baseline to reflect it) or reject it (triggering automatic restoration). The choice between automatic and approval-gated remediation is set at the blueprint and environment tier level.

Cost attribution is applied structurally at the point of provisioning, not as a manual tagging exercise. Every environment deployed through Self-Service is automatically tagged with the team, project, user, and business unit it belongs to. This metadata is carried through the entire environment lifecycle and is the basis for all cost reporting in Operate. Platform administrators see cost data across all teams and spaces. Team leads see cost data for their team. Individual users see cost data for their own environments. FinOps teams have access to structured cost reports by space, team, project, blueprint type, and user, with CSV export for external reporting.

Torque’s operational agents are not defined by two roles. SRE and FinOps are examples of the kinds of operational responsibilities agents are designed to cover, but the model extends to any area where informed, contextual decisions need to be made continuously and at scale, including security posture, compliance validation, capacity management, and more. What every agent shares is the same underlying design: they reason about the current state of what they are responsible for, evaluate the available options in context, and act within strictly defined governance boundaries. Each agent operates with a specific permission scope, a defined blast radius, and a full audit trail. An agent responsible for cost cannot modify infrastructure configurations. An agent responsible for remediation cannot make financial decisions. The boundaries are platform-enforced. Within them, the decision is the agent’s, not a condition in a script.

Yes, and production is where Operate is most valuable. The drift detection, cost attribution, approval workflows, and compliance audit trail are all designed to operate at the governance level required for production workloads. Production blueprints can be configured with stricter approval requirements, mandatory tag enforcement, and more conservative auto-remediation policies than development or staging environments. The operational dashboard gives platform teams a unified view across all environment types, with the ability to filter and act on each tier appropriately. Operate does not distinguish between pre-production and production, but it respects and enforces whatever policy distinctions the platform team defines.

A workflow is a decision tree. It evaluates conditions and selects from pre-defined branches. It is powerful for deterministic processes where every scenario can be anticipated and scripted in advance. But infrastructure operations are not fully deterministic. The same type of event in two different environments, at two different times, with two different histories, may warrant two different responses. A workflow cannot make that distinction because it was not written to. An agent can, because it reasons about context rather than matching conditions to branches. Torque’s SRE and FinOps agents evaluate the actual state of the environment, not just the triggering event. They consider what they know about the affected resource, the team that owns it, the history of similar issues, and the potential consequences of different responses. They then determine the appropriate action and execute it within the policy boundaries the platform team defined. Those boundaries are strictly enforced. Within them, the decision is the agent’s. This is what makes agentic operations genuinely different from sophisticated workflow automation — and why it handles real operational complexity in ways that workflows, however well-designed, fundamentally cannot.

Every action taken on every environment is logged automatically by the platform with full detail: who deployed it, who modified it, what changed and when, who approved or rejected out-of-band changes, who extended the TTL, and who destroyed it, all with precise timestamps. This log is generated by Operate as a structural output of normal operations, not as a separate compliance process that requires additional configuration. There is nothing to enable and nothing to maintain. When an auditor asks for the change history of a specific environment, the complete record is available immediately, in a structured, exportable format.