Agentic AI design patterns
Reusable answers to recurring problems, in two tracks: designing agentic systems, and building software with coding agents. Each pattern uses the same card: problem, diagram, when to use, trade-offs, a worked example and related patterns. Every pattern is tagged with the degrees of freedom it changes.
Agentic systems
Patterns for designing agents and multi-agent systems.
Core loop
How a single agent thinks, acts and checks itself.
Tool Use
The model calls external functions or APIs through structured schemas, grounding it in real data and actions.
Tool/action No. 02ReAct
Reason, act, observe, repeat. The default loop in most agent frameworks.
DecisionTool/action No. 03Plan-and-Execute
Write the plan first, then execute step by step and replan on failure.
Decision No. 04Reflection
The agent critiques and revises its own output before returning it.
Verification No. 05Prompt Chaining
A fixed sequence of LLM calls with checks between steps. Predictable and debuggable.
Decision No. 06Routing
Classify the request and send it to the right prompt, model or agent.
Decision No. 07Parallelization
Run independent subtasks at once, or run one task several times and vote.
CoordinationMulti-agent
How several agents divide and coordinate work.
Orchestrator–Workers
A lead agent decomposes the task at runtime, delegates, and merges results.
Coordination No. 09Evaluator–Optimizer
One agent generates, another grades against a rubric, and they loop until it passes.
VerificationCoordination No. 10Hierarchical Supervisor
Managers over leads over specialists, mirroring an org chart.
Coordination No. 11Handoff
Agents pass control and context peer-to-peer with no central boss.
Coordination No. 12Debate / Multi-Critic
Agents argue from different viewpoints and a judge picks the answer.
VerificationCoordination No. 13Blackboard
Agents read and write a shared workspace and contribute when they can.
CoordinationContextContext & memory
What the agent knows, remembers and forgets.
Agentic RAG
The agent decides when and what to retrieve, rewrites queries and checks relevance.
Context No. 15Tiered Memory
Working, episodic and semantic memory stored and recalled separately.
Context No. 16Context Compaction
Summarize or prune history so long tasks fit the context window.
Context No. 17Sub-agent Context Isolation
A sub-agent explores in a clean context and returns only a summary.
ContextCoordinationSafety & control
Where the agent's autonomy stops.
Human-in-the-Loop
Pause for human approval before risky or irreversible actions.
VerificationDecision No. 19Guardrails
Input and output filters, policy checks and schema validation around the agent.
Verification No. 20Bounded Autonomy
Limits enforced by policy, not prompts: tool allowlists, spend caps, scope.
Tool/actionDecision No. 21Sandboxed Execution
Code and actions run in isolation with checkpoints and rollback.
Tool/actionRuntime & interop
The platform the agent runs on and talks through.
Agent Harness
A reusable runtime for loop, tools, memory, permissions and progress, instead of one-off scripts.
Tool/actionContextDecision No. 23Code-as-Action
The agent writes code that chains many tool calls, instead of one JSON call per step.
Tool/action No. 24Skills / Progressive Disclosure
Packaged instructions load only when relevant, keeping base context small.
Context No. 25Protocol Interop (MCP / A2A)
Standard servers for tools and agent-to-agent messaging across vendors.
Tool/actionCoordination No. 26Budget-aware Planning
Each stage gets a token, time or dollar budget and degrades gracefully.
Decision No. 27Ambient Agents
Always-on agents watch event streams and act or alert without a prompt.
DecisionContext No. 28Generative UI
The agent renders or changes interface elements, not just chat text.
Tool/action No. 29Delegated Agent Identity
Agents get their own scoped, auditable credentials to act for a user.
Tool/actionEvaluation & learning
How the agent is graded and gets better.
Fast & slow
Pairing System 1 (fast, intuitive) and System 2 (deliberate) models.
Talker–Reasoner
A System 1 model keeps the conversation live while a System 2 model works in the background.
Coordination No. 33Fast/Slow Escalation
System 1 by default; escalate to System 2 when signals say the task is hard.
DecisionVerification No. 34Draft–Verify
System 1 drafts quickly; System 2 checks or corrects the draft.
Verification No. 35Respond-then-Refine
An instant first answer from System 1, followed by a refined System 2 version.
Decision No. 36Adaptive Thinking Budget
One model switches between no thinking and deep thinking based on difficulty.
Decision No. 37Metacognitive Monitor
A lightweight watcher spots surprises and switches the agent into System 2 mode.
Verification No. 38Habit Formation
Repeated System 2 solutions get compiled into fast System 1 skills or rules.
Context No. 39Reflex Layer
Instant System 1 responses for routine or safety-critical checks: triage, guardrails, stop/continue.
DecisionVerificationBuilding with coding agents
Patterns for developing software with AI coding agents.
Working agreements
What the repo and the team give the agent before work starts.
Spec-Driven Development
A short spec is the source of truth; the agent builds from it, not from chat history.
ContextDecision No. 41Project Memory File
A CLAUDE.md or AGENTS.md file holds conventions, commands and decisions that build up over time.
Context No. 42Agent-Ready Repo
One-command setup, scripted checks and clear docs, so an agent can run and verify its own work.
Tool/actionVerification No. 43Artifact Ownership
Each artifact is marked developer-owned, AI-drafted or joint, so everyone knows who decides what.
DecisionCoordinationBuild loop
How a coding agent moves from task to working code.
Plan-then-Code
The agent proposes a plan; a human approves it before any files change.
DecisionVerification No. 45Test-First Agent Loop
Tests define "done"; the agent iterates until they pass.
Verification No. 46Thin-Slice MVPs
Each session ships one small, working vertical slice (MVP1 → MVP2 → MVP3).
DecisionReview & scale
How agent changes land safely, and how to run several agents at once.
No patterns match these filters. Clear a filter or try a different word.