---
id: "action-design-hesitation"
type: "action-item"
source_timestamps: ["§ Designing from the Real Organization"]
tags: ["system-design", "risk-management"]
related: ["concept-professional-discretion", "quote-human-oversight-permanent"]
action: "Engineer confidence thresholds, anomaly detection, and escalation triggers into AI agents to simulate human hesitation."
outcome: "Prevention of machine-speed compounding errors and preservation of a safety buffer for edge cases."
speakers: ["K. Sudhir"]
sources: ["agentic"]
sourceVaultSlug: "hbr-seg-agentic"
originDay: 6
articleStem: "hbr-new-26-agentic-systems-implicit-rules"
sourceUrl: "https://hbr.org/2026/06/how-to-design-agentic-systems-around-the-implicit-rules-that-govern-your-company"
sourceTitle: "How to Design Agentic Systems Around the Implicit Rules that Govern Your Company"
---
# Design deliberate hesitation into AI agents

**Action:** Engineer confidence thresholds, anomaly detection, and escalation triggers into AI agents to simulate human hesitation.

**Outcome:** Prevention of machine-speed compounding errors and preservation of a safety buffer for edge cases.

Because AI agents do not naturally pause when something feels wrong (they lack [[concept-professional-discretion]]), you must *engineer hesitation into the system*. Build **confidence thresholds, anomaly detection, and escalation triggers.** Treat human oversight as a **permanent design feature** — not a temporary phase — to handle the unanticipated risks that programmed hesitation misses ([[quote-human-oversight-permanent]], [[contrarian-human-oversight-permanent]]).

This is Step 2 of [[framework-design-real-organization]] and directly guards against [[concept-machine-speed-compounding]] and repeats of the [[entity-air-canada-d6|Air Canada]] failure. Requires familiarity with human-in-the-loop architecture ([[prereq-hitl-concepts]]).


## Related across articles
- [[concept-independent-verification-safeguards]]
- [[concept-human-in-the-loop-escalation]]
- [[action-define-decision-rights]]