An agent control plane must prove effective authority, cost and recovery
TCS says it is building a governance layer for its own, client and third-party agents. Buyers should test the effective path across vendors instead of treating one dashboard as control.
What happened
TCS’s COO told Financial Express that the company is building a control plane to orchestrate agents from TCS, clients and third parties across enterprise applications while managing governance and cost.
Why it matters
A central layer can improve visibility, but heterogeneous agents retain different identities, permissions, memories and failure modes. A control-plane claim becomes useful only when tested along the complete action path.
Financial Express reported on 12 October that TCS is building a governance layer for agents developed by the company, customers and third parties across enterprise applications. COO Aarthi Subramanian described it as a control plane for security, deployment and cost. The report says TCS’s AI business reached $3 billion in annualised revenue, but that figure is not evidence that the control layer works.
TCS’s BPS 2030 paper proposes reusable capabilities with a human control tower overseeing confidence, exceptions, service levels, approvals and compliance. This is a vendor architecture and future operating model, not an independent evaluation.
Test the effective path
Create one cross-vendor scenario with different user roles. Trace the initiating identity, delegated credentials, model and agent versions, tool calls, data disclosed, approvals, side effects, cost and rollback. Revoke a permission in the source system and verify every cached, delegated and third-party path fails within a declared objective.
Then inject a cost spike, stale memory, conflicting policy and unavailable vendor. Confirm that the control plane can stop the action, preserve evidence and restore work without silently broadening authority. Measure unobserved calls, policy mismatches, time to revoke, recovery time and total cost per validated outcome.
Run the same test without the control plane. A central layer earns its place only if it improves containment and evidence enough to justify another dependency and failure domain. Record outages and manual fallback, not only successful orchestration.
The immediate decision is not whether a unified dashboard exists. It is whether the organisation can prove that control survives every boundary the agent crosses.