Skip to main content

Paradigm Selection Playbook

A Practitioner's Guide to Choosing the Right Agentic AI Pattern​

Document ID: A1-PARADIGM-PLAYBOOK
Version: 1.0
Category: P1 - Implementation Guides

Executive Summary​

This playbook provides a systematic approach to selecting the optimal agentic AI paradigm for specific use cases. Based on analysis of 400+ research papers, we present decision frameworks, scenario mappings, and implementation checklists.

The Four Paradigms​

ParadigmKnowledge SourceAgency TypeBest ForRisk Profile
LSR (Latent Space Reasoner)Implicit (parametric)CognitionCreative synthesis, brainstormingHallucination risk
GS (Grounded Synthesizer)Explicit (retrieval)CognitionResearch, analysis, fact-basedRetrieval bottlenecks
EP (Emergent Planner)Implicit (parametric)ExecutionNovel problem-solving, adaptiveAction hallucination
VE (Verifiable Executor)Explicit (protocols)ExecutionCompliance, regulated processesRigidity

Primary Decision Tree​

Q1: Is full auditability and compliance critical?
├─► YES → Q2: Does task require taking actions?
│   ├─► YES → Q3: Is there a defined process?
│   │   ├─► YES → ✓ VE (Verifiable Executor)
│   │   └─► NO → ✓ Hybrid: EP with audit trail
│   └─► NO → ✓ GS (Grounded Synthesizer)
└─► NO → Q4: Is factual accuracy critical?
    ├─► YES → ✓ GS (Grounded Synthesizer)
    └─► NO → ✓ LSR (Latent Space Reasoner)

Scenario-Based Selection​

Content & Communication​

ScenarioRecommendedRationale
Marketing copyLSRCreative synthesis
Technical docsGSAccuracy, citations
Legal correspondenceVETemplates, compliance
Blog postsLSR→GSCreative draft, fact-check

Research & Analysis​

ScenarioRecommendedRationale
Market researchGSMulti-source, citations
Due diligenceGS+VEEvidence + audit trail
BrainstormingLSRCreative exploration
Strategic planningEPMulti-step reasoning

Operations & Workflow​

ScenarioRecommendedRationale
Invoice processingVEDefined workflow
Customer onboardingVESequential validation
Exception handlingEPNovel situations
IT ticket routingVEClassification rules

Implementation Checklists​

LSR Checklist​

  • Use case tolerates occasional inaccuracies
  • Creative/generative output valued
  • Human review part of workflow
  • Self-consistency sampling (3-5 paths)

GS Checklist​

  • Knowledge base available
  • Citation/attribution required
  • Vector store configured
  • Relevance threshold set (0.7+)

EP Checklist​

  • Task requires multi-step execution
  • Environment state observable
  • Iteration budget defined
  • Reflexion memory enabled

VE Checklist​

  • Defined protocols exist
  • Compliance/audit required
  • Audit logger configured
  • Rollback capability tested

Hybrid Pattern Recipes​

Creative-to-Verified (LSR→GS→VE)​

  1. LSR: Generate creative draft
  2. GS: Fact-check, add citations
  3. VE: Apply compliance rules

Research-to-Action (GS→EP)​

  1. GS: Gather and synthesize information
  2. EP: Plan actions based on findings
  3. EP: Execute with adaptation

Compliant Autonomy (EP+VE)​

  1. EP: Plan adaptive approach
  2. VE: Validate each step against protocols
  3. EP: Adapt based on validation feedback

Paradigm Selection Scorecard​

FactorWeightLSRGSEPVE
Creativity needed0.155231
Factual accuracy0.201534
Multi-step execution0.151254
Audit/compliance0.201325
Novel scenarios0.104251
Latency sensitivity0.105314
Cost sensitivity0.104313

Interpretation: Score > 4.0 = Strong fit | 3.0-4.0 = Good fit | 2.0-3.0 = Consider hybrid | < 2.0 = Poor fit

Anti-Patterns​

Anti-PatternProblemSolution
LSR for ComplianceInconsistent, no auditSwitch to VE
VE for Novel ScenariosHigh exception ratesAdd EP layer
GS without ThresholdsIrrelevant citationsSet 0.7+ threshold
EP without LimitsRunaway costsIteration budgets
Single Paradigm for ComplexPoor subtask fitDesign hybrids

Document maintained by CODITECT Architecture Team