NPC Behavior Workbench FSM vs Agentic AI

Comparing Deterministic Scripted State Trees with Generative Utility & Memory Agents

Greed: 0.85

Scripted FSM NPC Engine

State: PATROL
Current FSM Transition Logic

IF Noise within vision range -> INVESTIGATE
IF Player detected -> ALERT
ELSE -> PATROL (Ignores dropped items)

FSM Execution Metrics
Tick Latency: 0.02 ms
Token Cost: 0 tokens
Predictability: 100% Deterministic

Memory & Utility AI Agent

Dynamic Action: EVALUATING
Agent Epistemic Memory Trail 0 items
  • Initializing perceptual field...
Utility Decision Vectors
Utility Duty: 0.40
Utility Greed: 0.85
Inferred Prompt Latency: 115 ms

Live System Telemetry

Emergent Behavior Index: HIGH (0.88)
State Hallucination Risk: LOW (4.2%)
Est. 1k Ticks Token Cost: $0.00 vs $0.42

Calculated via 120-token context vector per agent memory evaluation cycle.

Emergent Narrative & Divergence Log

Comparing NPC Decision Offsets
💡

Observed Divergence: While the traditional FSM strictly ignores unhandled environmental objects like dropped gold coins unless explicitly coded in state rules, the Memory & Utility AI Agent perceives the coin, updates its epistemic memory vector, and dynamically overrides patrol duty to steal the gold coin first.

Enjoy this tool? Build your own with Super