Agentic Paradigms Quick Reference Card

The Two Fundamental Trade-offs

TRADE-OFF 1: RELIABILITY ← → CREATIVITY
├── IMPLICIT KNOWLEDGE (Creative)
│   └── Model parameters as source of truth
│   └── Risk: Hallucination
└── EXPLICIT KNOWLEDGE (Reliable)
    └── External sources as source of truth
    └── Risk: Rigidity

TRADE-OFF 2: SAFETY ← → AUTONOMY
├── EVENT COGNITION (Safe)
│   └── Understand and advise
│   └── Human makes decisions
└── GOAL EXECUTION (Autonomous)
    └── Plan and execute
    └── Agent takes action

The Four Paradigms at a Glance

Paradigm	Knowledge	Agency	One-Liner
Latent Space Clinician (LSC)	Implicit	Cognition	Creative synthesis from internal knowledge
Grounded Synthesizer (GS)	Explicit	Cognition	Evidence-based information synthesis
Emergent Planner (EP)	Implicit	Execution	Autonomous multi-step planning
Verifiable Workflow Automator (VWA)	Explicit	Execution	Protocol-driven process execution

Five Technical Components

1. Strategic Planning

Component	LSC	GS	EP	VWA
Decomposition	Chain-of-thought	Query dependency graph	Cognitive maps	Protocol mapping
Iteration	Self-consistency	Gap-filling loops	Self-reflection	Closed-loop correction

2. Memory Management

Component	LSC	GS	EP	VWA
Parametric	Knowledge source	Query generator	Strategic engine	Intent parser
Non-Parametric	Context window	Evidence ledger	Workflow log	State register

3. Action Execution (Explicit paradigms only)

Action Type	GS Purpose	VWA Purpose
Knowledge-Based	Logical grounding via KG	Verifiable inference via DB
Search-Based	Forage-Constrain-Attribute	Optimize-Route-Validate
Tool-Use	Liability offloading	Clinical orchestration

4. Collaboration

Topology	Description	Best For
Single-Agent	One model handles all	Simple tasks
Dominant (Hierarchical)	Central orchestrator delegates	Complex workflows
Distributed (Peer-to-Peer)	Agents negotiate consensus	Ambiguous decisions

5. Evolution

Paradigm	What Evolves	How
LSC	Model parameters	Continual learning
GS	Query strategies	Efficiency optimization
EP	Action policies	RL / meta-learning
VWA	Workflow components	Evolutionary refinement

Decision Matrix

When to Use Each Paradigm

                          HIGH STAKES?
                    NO                YES
              ┌─────────────────┬─────────────────┐
              │                 │                 │
   NEEDS      │      LSC        │       GS        │
   CREATIVITY │  (Creative +    │  (Reliable +    │
              │   Advisory)     │   Auditable)    │
              │                 │                 │
              ├─────────────────┼─────────────────┤
              │                 │                 │
   NEEDS      │      EP         │      VWA        │
   EXECUTION  │  (Flexible +    │  (Predictable + │
              │   Autonomous)   │   Compliant)    │
              │                 │                 │
              └─────────────────┴─────────────────┘

Key Patterns Cheat Sheet

LSC Pattern: Chain-of-Thought + Self-Consistency

Query → Multiple reasoning paths → Vote → Consensus answer

GS Pattern: Decompose + Ground + Cite

Query → Sub-queries → Retrieve evidence → Synthesize with citations

EP Pattern: Plan + Execute + Reflect + Adapt

Goal → Cognitive map → Execute step → Reflect → Adjust plan → Repeat

VWA Pattern: Map + Validate + Execute + Log

Request → Match protocol → Validate preconditions → Execute → Audit trail

Risk Mitigation Quick Reference

Risk	LSC Mitigation	GS Mitigation	EP Mitigation	VWA Mitigation
Hallucination	Human validation	Evidence firewall	Reflection checkpoints	Protocol constraints
Staleness	Periodic fine-tuning	Source updates	Policy updates	Workflow versioning
Cascade Failure	N/A (advisory)	Source diversity	Rollback capability	Circuit breakers
Compliance	Not recommended	Audit trails	Human escalation	Built-in logging

Enterprise Application Mapping

Use Case	Primary Paradigm	Supporting Paradigm
Document Generation	GS	LSC (creativity)
Data Entry Automation	VWA	-
Customer Support	GS	LSC (empathy)
Process Automation	VWA	EP (exceptions)
Research Synthesis	GS	LSC (insights)
Strategic Analysis	LSC	GS (grounding)
Compliance Reporting	VWA	GS (verification)
Exception Handling	EP	VWA (escalation)

Memory Architecture Reference

┌─────────────────────────────────────────────────────────────┐
│                    MEMORY STACK                             │
├─────────────────────────────────────────────────────────────┤
│  STATIC LAYER                                               │
│  └── Business rules, protocols, domain knowledge            │
├─────────────────────────────────────────────────────────────┤
│  DYNAMIC LAYER                                              │
│  └── Session state, workflow progress, entity tracking      │
├─────────────────────────────────────────────────────────────┤
│  AUDIT LAYER                                                │
│  └── Action logs, decision traces, compliance records       │
├─────────────────────────────────────────────────────────────┤
│  LEARNING LAYER                                             │
│  └── Pattern cache, optimization data, feedback history     │
└─────────────────────────────────────────────────────────────┘

Collaboration Topology Selection

Single Agent ← Choose when →
  ✓ Task is simple and well-defined
  ✓ Low complexity
  ✓ Resource constraints

Dominant Topology ← Choose when →
  ✓ Clear hierarchy needed
  ✓ Complex multi-step workflows
  ✓ Centralized control preferred

Distributed Topology ← Choose when →
  ✓ Multiple perspectives needed
  ✓ Ambiguous decisions
  ✓ Peer review beneficial

Metrics Reference

Category	Key Metrics
Accuracy	Completion rate, Error rate, Compliance score
Efficiency	Latency, Human touches per task, Cycle time
Quality	Citation accuracy, Reasoning validity, User satisfaction
Safety	Escalation rate, Rollback frequency, Audit findings

Implementation Checklist

For Any Paradigm

Define clear component boundaries (Planning, Memory, Action, etc.)
Implement appropriate memory architecture
Establish evolution mechanism
Create audit logging
Build human escalation paths

For Explicit Paradigms (GS, VWA)

Define knowledge sources
Implement evidence firewalling
Create citation/attribution system
Validate source freshness

For Execution Paradigms (EP, VWA)

Define rollback mechanisms
Implement circuit breakers
Create state persistence
Build monitoring dashboards

Source Reference

Paper: "Reinventing Clinical Dialogue: Agentic Paradigms for LLM-Enabled Healthcare Communication"
Authors: Zhi, Zhao, Wu, Zhao, Zhu (Tianjin University, Chinese Academy of Sciences)
Scope: 300+ papers surveyed | Dec 2025
Primary Contribution: Two-axis taxonomy for agentic systems (Knowledge Source × Agency Objective)

The Two Fundamental Trade-offs​

The Four Paradigms at a Glance​

Five Technical Components​

1. Strategic Planning​

2. Memory Management​

3. Action Execution (Explicit paradigms only)​

4. Collaboration​

5. Evolution​

Decision Matrix​

When to Use Each Paradigm​

Key Patterns Cheat Sheet​

LSC Pattern: Chain-of-Thought + Self-Consistency​

GS Pattern: Decompose + Ground + Cite​

EP Pattern: Plan + Execute + Reflect + Adapt​

VWA Pattern: Map + Validate + Execute + Log​

Risk Mitigation Quick Reference​

Enterprise Application Mapping​

Memory Architecture Reference​

Collaboration Topology Selection​

Metrics Reference​

Implementation Checklist​

For Any Paradigm​

For Explicit Paradigms (GS, VWA)​

For Execution Paradigms (EP, VWA)​

Source Reference​