Engineer Curates 15 AI Agent Design Patterns Across Single-Agent, Multi-Agent and Feedback Tiers
MaryamMiradi · x · 2026-11-12
X user @MaryamMiradi, citing experience building 400+ AI agents, curated 15 key agent design patterns in three tiers:
Tier 1: Single-agent patterns
- ReAct: alternate reasoning and acting so the agent thinks before each tool call.
- Plan-and-Execute: generate a full plan upfront, then execute step by step.
- Reflection / Self-Critique: the agent reviews and iterates on its own output.
- Tool Use / Function Calling: the agent decides which external tool to invoke and when.
Tier 2: Multi-agent orchestration
- Orchestrator-Subagent: a coordinator decomposes goals and delegates to specialized workers.
- Supervisor: routes tasks, monitors outputs and enforces quality gates.
- Hierarchical Agents: orchestrators manage other orchestrators across levels.
- Sequential Pipeline: each agent completes a task and passes results on.
- Parallel Fan-Out / Fan-In: split tasks across agents, then merge results.
- MapReduce: distribute subtasks and aggregate into one output.
- Debate / Adversarial: agents argue opposing positions and a judge resolves.
Tier 3: Iterative and feedback loops
- Evaluator-Optimizer: one agent generates, another scores, looping until quality is met.
- Critic-Actor: a critic gives structured feedback and the actor refines.
- Self-Healing / Retry Loop: diagnose failures and retry with a corrected strategy.
- HITL: humans approve, correct or redirect at defined checkpoints.
The author frames single-agent patterns as what one agent can do, multi-agent patterns as coordination under load, and feedback loops as recovery and improvement, while promoting a paid agent training course.
More from coding & agent
- Continual-ARC benchmark: leading AI agents fail to learn from experience — nikola_mr64990 · 2026-11-12
- Rayline launches ARC, a per-turn model router claiming 90% cost cuts — testingcatalog · 2026-11-12
- Karpathy's llm-wiki: a pattern for building personal knowledge bases with LLM agents — stochasticchasm · 2026-11-12
- Agent self-improvement with frozen weights: failure traces add just 4pp, blind search gets 60% — burny_tech · 2026-11-12
- Microsoft on agentic schedulers: LLM proposes, deterministic code validates — WirelessLife · 2026-11-12
- Visual block scripting drives an AI opponent in a classic arcade game — ezshine · 2026-11-12