ReAct (Reason + Act)
ReAct is a prompting and agent-design paradigm that interleaves reasoning (chain-of-thought) with acting (tool use or environment interaction). Introduced by Yao et al. (2022), it gives language models a structured loop -- Thought, Action, Observation -- that mirrors how humans solve problems: think about what to do, do it, observe the result, and repeat.
Chain-of-Thought Reasoning
Chain-of-Thought (CoT) prompting encourages a language model to produce intermediate reasoning steps before arriving at a final answer. Instead of jumping straight to the output, the model "shows its work," which dramatically improves performance on arithmetic, commonsense, and symbolic reasoning tasks.
Tool Use Pattern
Tool use is the ability of an LLM-based agent to select, invoke, and interpret the results of external tools at runtime. It transforms a language model from a text generator into an actor that can query databases, call APIs, run code, and manipulate files. This pattern is foundational to virtually every production agent.
Reflection and Self-Critique
Reflection is a design pattern where an agent evaluates its own output, identifies flaws, and uses that critique to improve subsequent attempts. Instead of accepting the first response as final, the agent enters a deliberate self-improvement loop. This pattern is central to frameworks like Reflexion and is one of the most effective ways to boost agent quality without changing the underlying model.
Multi-Agent Collaboration
Multi-agent systems use two or more specialized LLM agents that work together to solve tasks no single agent handles well alone. Each agent has a distinct role, persona, or capability, and they coordinate through structured communication patterns. This mirrors how human teams work: specialists collaborate, debate, and delegate to produce better outcomes than any individual could achieve.
Plan-and-Execute
The Plan-and-Execute pattern separates agent behavior into two distinct phases: first create a plan (a sequence of steps), then execute each step one at a time. After execution, the agent may revise the plan based on what it learned. This separation makes complex tasks more manageable, debuggable, and controllable than interleaved approaches like ReAct.
Self-Refinement
Self-refinement is a pattern where an agent generates an initial output, critiques it, and iteratively revises it until a quality threshold is met or a budget is exhausted. Unlike the broader Reflection pattern (which emphasizes cross-episode memory and frameworks like Reflexion), self-refinement focuses specifically on within-task output polishing -- taking a single artifact from draft quality to production quality.
Human-in-the-Loop
Human-in-the-Loop (HITL) is a design pattern where an AI agent pauses at critical decision points to request human review, approval, or input before proceeding. Rather than fully autonomous operation, the agent operates on a spectrum -- handling routine tasks independently while escalating high-stakes, ambiguous, or low-confidence situations to a human operator.