liveBook · Manning

Simulation

As explained in Architecting Autonomous Agentic Systems

Simulation is a technique that models or rehearses system behavior across varied inputs, conditions, and scenarios to evaluate outcomes and risks.

When an autonomous clinical pathway agent operates under uncertain model confidence, simulation rehearses how its decisions and escalation behavior change before deployment. It replaces confidence based only on fixed design artifacts with evidence from varied, plausible conditions. section 2.2.3

Simulation treats behavior as probabilistic and context-dependent rather than assuming that all relevant paths can be specified and tested in advance. Its architectural purpose is to explore decision boundaries, escalation thresholds, and behavior under stress—not to prove model accuracy or component correctness. section 2.2.3

Use simulation to:

  • Rehearse failure modes and boundary conditions across a range of plausible scenarios. section 2.2.3
  • Expose escalation thresholds that permit questionable decisions before authority returns to humans. section 2.2.3 section 2.4
  • Support a resilience loop linking autonomy boundaries, observability, policy enforcement, and escalation. section 2.4

Clinical pathway model drift

A clinical pathway agent was tested against gradual confidence degradation from 0.12 to 0.20 to 0.30, sudden degradation from 0.12 to 0.35, and simultaneous drift across multiple models. The simulations showed that a 0.35 escalation threshold was too permissive because questionable decisions had already occurred. Lowering the threshold to 0.30 tightened the autonomy boundary before deployment. section 2.2.3

How Architecting Autonomous Agentic Systems covers it

How deep it goes
Treats simulation architecturally through failure modes, boundary conditions, stress behavior, and model drift, stopping short of exhaustive correctness.
Examples
Uses clinical pathway model-drift scenarios with gradual, sudden, and simultaneous confidence-score degradation.
What it assumes
Assumes awareness of autonomous-system uncertainty, probabilistic behavior, and decision boundaries rather than deterministic paths.
What you can do after
Evaluate uncertainty-driven behavior and adjust escalation thresholds before deploying autonomous systems.
How it connects
Connects simulation with policy-as-code, orchestration, autonomy boundaries, observability, and escalation triggers.

Also explained in

Compare how 11 books explain simulation →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Architecting Autonomous Agentic Systems. The section links point to the parts of the book it draws on.