Agentic Pattern Catalog

Agentic Pattern Catalog

Maintained by: Ada · Agentic Pattern Designer Last updated: 2026-03-17

Overview

This catalog contains the reusable agentic patterns used across Talent Factory production lines to build digital talents. Each pattern is documented with when-to-use guidance, mechanics, examples, and tradeoffs.

Total patterns: 21

Sources:

  • Internal usage — patterns already used implicitly in the factory
  • FOR-0012 — "Master ALL 20 Agentic AI Design Patterns" (Google engineer's book summary)
  • garrytan/gstack — Cherry-picked patterns from Garry Tan's founder toolkit (RD-0013)

Orchestration

Patterns for how multiple agents coordinate, delegate, and share work.

Pattern File Status Summary
Prompt Chaining prompt-chaining.md Active Break a task into sequential steps where each output feeds the next
Orchestrator-Workers orchestrator-workers.md Active Central agent coordinates specialist workers and synthesizes results
Routing routing.md Active Analyze requests and direct them to the right specialist agent
Parallelization parallelization.md Documented Split independent subtasks across parallel workers for throughput
Multi-Agent Coordination multi-agent-coordination.md Active Multiple specialized agents collaborate via shared resources and protocols
Quality Gates quality-gates.md Active Checkpoints between pipeline stages that enforce minimum quality standards
Control Plane Separation control-plane-separation.md Active Separate coordination (who/when/cost) from execution (skills/knowledge) using an orchestration platform

Future candidates: swarm, map-reduce, hierarchical delegation

Tool Use

Patterns for how agents interact with external tools, APIs, and services.

Pattern File Status Summary
Tool Use tool-use.md Active Select, invoke, and parse external tools (file I/O, APIs, CLI)
Fallback and Recovery fallback-and-recovery.md Documented Degrade gracefully when operations fail — retry, alternative, or escalate
Guardrails and Safety guardrails-and-safety.md Documented Validate inputs and outputs against safety, compliance, and brand rules

Future candidates: structured-output formatting, tool selection routing

Memory

Patterns for how agents persist, retrieve, and share context.

Pattern File Status Summary
Memory Management memory-management.md Documented Classify, store, and retrieve information across sessions (short-term, episodic, long-term)
Retrieval-Augmented Generation retrieval-augmented-generation.md Documented Ground responses in a searchable knowledge base to reduce hallucination
Shared State shared-state.md Active Multiple agents read/write a common data store for coordination

Future candidates: summary compaction, conversation buffer windowing

Reasoning

Patterns for how agents think through problems and make decisions.

Pattern File Status Summary
Evaluator-Optimizer Loop evaluator-optimizer-loop.md Active Iteratively generate, evaluate, and improve until quality threshold is met
Reflection reflection.md Documented Self-critique a draft against quality standards, then revise
Plan-Then-Execute plan-then-execute.md Documented Create a structured plan before executing, with step-by-step tracking

Future candidates: chain-of-thought prompting, tree-of-thought search, adversarial debate

Human-in-the-Loop

Patterns for how agents involve humans in decisions and workflows.

Pattern File Status Summary
Approval Gate approval-gate.md Active Pause for explicit human approval before proceeding on high-stakes decisions
Escalation escalation.md Documented Route situations the agent cannot handle confidently to a human or higher authority
Feedback Loop feedback-loop.md Active Collect user feedback and apply corrections to improve system behavior over time

Future candidates: confidence threshold routing, human override

Quality Assurance

Patterns for validating, testing, and verifying deliverables and deployments.

Pattern File Status Summary
Browser-Based QA browser-based-qa.md Cataloged Real browser testing with Playwright for visual and interactive validation
Cross-Model Review cross-model-review.md Cataloged Independent second-model adversarial review to catch blind spots
Security Threat Review security-threat-review.md Cataloged OWASP Top 10 + STRIDE threat modeling with confidence filtering
Canary Monitoring canary-monitoring.md Cataloged Post-deploy health checks, performance regression detection, soak period
Design Dimension Scoring design-dimension-scoring.md Cataloged Structured 0-10 scoring across design quality dimensions with gap remediation

Source: garrytan/gstack (RD-0013) — cherry-picked and adapted for factory use

How to Add a Pattern

Use the /pattern-catalog add command, or manually create a pattern file following the standard template:

departments/it-architecture/agentic-pattern-designer-ada/patterns/{category}/{pattern-name}.md

Every pattern must include: name, category, when-to-use, how-it-works, example, tradeoffs.

After adding the file, update this index with the new entry in the appropriate category table.