Notifications
Clear all
Topic starter
16/09/2026 9:21 am
TL;DR: Supporting thousands of cybersecurity agents requires a shared control plane for identity, policy, versioning, and scoped execution so each run stays inside customer boundaries and remains auditable, according to Cogent. The key lesson is that agent governance breaks unless authority, version, and approval boundaries are enforced outside the model and tied to every run.
NHIMG editorial — based on content published by Cogent: Blog Engineering Thousands of Agents: How We Build and Manage an Agent Workforce
Questions worth separating out
Q: How should security teams govern AI agents that can access enterprise systems?
A: Security teams should govern AI agents as non-human identities with explicit ownership, scoped privileges, and continuous monitoring.
Q: What breaks when agent permissions are not tied to a specific version?
A: Auditability breaks first, followed by accountability.
Q: How do organisations decide when an autonomous agent needs approval versus direct execution?
A: Use the sensitivity of the action, not the sophistication of the model, to decide.
Practitioner guidance
- Separate identity from model behaviour Assign every agent an immutable identity, version, and scope record so review can answer who acted, under which definition, and with what authority.
- Move approval decisions outside the model Use an external policy engine or gateway to allow, escalate, or deny high-impact actions before execution.
- Treat agent definitions as governed lifecycle objects Track prompt blocks, tools, skills, and permissions as versioned assets with release history and rollback paths.
What's in the full article
Cogent's full blog post covers the operational detail this post intentionally leaves for the source:
- The agent development platform structure for versioning, deployment, and execution across thousands of agent definitions.
- The policy engine and gateway flow that evaluates allow, escalate, and deny decisions outside the model.
- The sandbox control plane logic for credentials, quotas, network rules, and orphaned environment cleanup.
- The trace and evaluation model that links every run to evidence, policy decisions, and the exact agent version.
👉 Read Cogent's blog post on building and managing an agent workforce →
Agent workforce control planes: what identity teams need to re-evaluate?
Explore further
16/09/2026 9:22 am
Agent workforce governance should be treated as NHI lifecycle management with runtime evidence, not as model operations. The article shows that agent creation, version selection, policy enforcement, execution, and traceability all belong in one governed chain. That is the same lifecycle logic identity teams already apply to service accounts and privileged workloads, but with faster iteration and more frequent behavioural change. Practitioners should stop treating agent orchestration as a separate discipline from identity governance.
A few things that frame the scale:
- The average estimated time to remediate a leaked secret is 27 days, despite 75% of organisations expressing strong confidence in their secrets management capabilities, according to The State of Secrets in AppSec.
- Only 44% of developers are reported to follow security best practices for secrets management, exposing a significant developer behaviour gap.
A question worth separating out:
Q: Why do traces matter more than final outputs for agent oversight?
A: Final outputs tell you whether the answer looked right, but traces show where the agent’s reasoning, retrieval, or tool use broke down. That distinction matters when the failure is caused by missed context, the wrong data source, or premature stopping. Traces turn opaque behaviour into evidence that can be reviewed and improved.
👉 Read our full editorial: Agent workforce governance needs versioned control planes, not ad hoc wrappers
