Topic starter
01/08/2026 10:50 pm
TL;DR: Agentic workflows that rely only on explicit rules fail when context lives in human judgment, not in logs or policy docs, according to Tonic. The practical problem is not whether AI can act, but when it should stop and escalate before a correct decision becomes an operational loss.
Editorial analysis by NHI Mgmt Group, based on content published by Tonic: “On Building Agents & Agentic Workflows (Part 4)”.
Key questions
Q: Where do AI agents fail when explicit rules are not enough?
A: They fail at the point where the environment contains context that the policy model cannot express, such as operational exceptions, business timing, or local system knowledge.
Q: Why should organisations route agent exceptions to humans?
A: Because some enterprise decisions depend on tacit context that cannot be reduced to a rule set.
Q: How can security teams spot when an AI agent is overconfident?
A: Look for cases where the agent acts on a policy signal without recognising a local exception, especially in high-impact workflows.
Practitioner guidance
- Define escalation thresholds for agent collisions Map the decision points where an AI agent must stop because policies conflict, context is missing, or business impact is uncertain.
- Capture tacit exceptions as governed decision patterns Document recurring human override patterns such as maintenance windows, business-critical systems, and local resilience assumptions.
- Separate permission from judgment in agent workflows Authorize the agent to act only within a clear action scope, then require human judgment when the system cannot reconcile security, continuity, or operational priorities.
Bottom line: Tacit human knowledge is not a soft skill in agentic systems, it is a control dependency that explicit rules cannot replace.
Explore further
View Full Forum → | NHI Foundation Course → | Our Services → | Read the full analysis →
01/08/2026 10:58 pm
Tacit knowledge is now part of the control plane for AI agents. The article is right to treat human judgment as a design element rather than a fallback. When an agent cannot represent the full decision context, the issue is governance completeness, not model intelligence. For identity teams, that means delegated action requires a stopping condition that is as intentional as the permission to act.
A few things that frame the scale:
- AI-related credential leaks surged 81.5% year-over-year in 2025, with the surrounding AI infrastructure leaking 5x faster than core LLM providers, according to the State of Secrets Sprawl 2026.
A question worth separating out:
Q: What should teams do when agent autonomy and business continuity conflict?
A: They should define a collision policy that names who decides, what evidence the agent must present, and which actions remain blocked until review completes. This is a governance question, not a model tuning problem. The objective is to preserve continuity without letting the agent make a locally correct but globally harmful choice.
👉 Read our full editorial: Tacit knowledge is now a control plane issue for AI agents