Concept
Reflex Agent
The simplest type of AI agent - one that maps directly from the current observation to an action, without any planning or memory of past states.
Added May 18, 2026
Agent architectures exist on a spectrum from simple to complex. At the simplest end is the reflex agent - a system that takes the current input and immediately produces an output based on a fixed set of rules or a direct learned mapping, without considering past history or thinking ahead about consequences.
The name comes from the analogy to biological reflexes: a reflex happens automatically in response to a stimulus, without involving conscious deliberation. When you pull your hand from a hot surface, you do not think about it - the action happens directly from the sensation. A reflex agent operates the same way: stimulus in, response out, with no reasoning in between.
In the context of AI agents, a pure reflex agent might be a chatbot that has a fixed response to each category of input - "if the input contains the word cancel, respond with the cancellation policy." Or a simple game-playing agent that maps each game state directly to an action according to a lookup table. No planning, no memory, no consideration of what happened before or what might happen next.
Reflex agents are useful in domains where the right action can be determined from the current observation alone - where no history or planning is needed. Reactive game opponents, simple alert systems, and deterministic decision systems can all be implemented as reflex agents effectively.
The limitations appear quickly in more complex domains. A reflex agent cannot handle situations where the same observation requires different responses depending on history ("is this the first time the user asked this question, or the third?"). It cannot plan ahead to avoid future problems. It cannot accumulate information across multiple observations to build a more complete picture.
Understanding reflex agents is useful precisely because they illustrate the minimum - and thus clarify what additional capabilities (memory, planning, learning) more sophisticated agents add, and why those additions matter for handling real-world complexity.
Analogy
A thermostat. When the temperature drops below the set point, it turns on the heater. When it rises above, it turns off. No memory of how long the heater has been running, no prediction of what the temperature will be in an hour, no consideration of energy costs - just: current state, fixed response. The thermostat is the prototypical reflex agent.
Real-world example
Early customer service chatbots were reflex agents: they matched keywords in user messages to canned responses. 'Return' triggers the return policy response. 'Billing' triggers the billing FAQ. These systems worked for common, simple queries but immediately failed when customers asked something that did not match a predefined pattern. Their lack of memory or planning meant they could not handle multi-turn conversations coherently.
Why it matters
Reflex agents serve as the conceptual baseline in agent design. They illustrate the minimum viable agent and clarify what memory, planning, and learning capabilities add in practice. Most real AI agents are combinations of reactive and deliberative behaviour - quick reflexes for common situations, deliberate reasoning for novel or complex ones. Understanding reflex agents is understanding one end of that spectrum.
In the news
No recent coverage - check back later.
Related concepts