Skip to main content

📘 Components of AI agents

AI agents have four main components: perception, planning and reasoning, tools, and memory.

Perception​

Perception, in the context of AI agents, is the mechanism by which the agent gathers information about its environment. Text inputs are currently the most common perception mechanism for AI agents, but we are slowly progressing towards audio, visual, multimodal or even physical sensory inputs.

Planning and reasoning​

AI agents use user prompts, self-prompting and feedback loops to break down complex tasks, reason through their execution plan and refine it as needed.

Some common design patterns for planning and reasoning in AI agents are as follows:

Chain of Thought (Cot) Prompting​

In this approach, the LLM is prompted to generate a step-by-step explanation or reasoning process for a given task or problem.

Here is an example of a zero-shot CoT prompt:

Given a question, write out in a step-by-step manner your reasoning for how you will solve the problem to be sure that your conclusion is correct. Avoid simply stating the correct answer at the outset.

ReAct (Reason + Act)​

In this approach, the LLM is prompted to generate reasoning traces and task-specific actions in an interleaved manner, allowing for greater synergy between the two: reasoning traces help the model induce, track, and update action plans, while actions allow it to interface with external sources or tools, to gather additional information.

Here is an example of a ReAct prompt:

Answer the following questions as best you can. You have access to the following tools:{tools}
##
Use the following format:
Question: the input question you must answer
Thought: you should always think about what to do
Action: the action to take, should be one of [{tool_names}]
Action Input: the input to the action
Observation: the result of the action
... (this Thought/Action/Action Input/Observation can repeat N times)
Thought: I now know the final answer
Final Answer: the final answer to the original input question

Reflection​

Reflection involves prompting an LLM to reflect on and critique past actions, sometimes incorporating additional external information such as tool observations. The generation-reflection loop is run several times before returning the final response to the user. Reflection trades a bit of extra compute for a shot at better output quality.

Tools​

Tools are interfaces for AI agents to interact with the external world in order to achieve their objectives. These can be APIs, vector databases, or even specialized machine learning models.

Memory​

The memory component allows AI agents to store and recall past conversations, enabling them to learn from these interactions.

There are two main types of memory for AI agents:

  • Short-term memory: Stores and retrieves information from a specific conversation.

  • Long-term memory: Stores, retrieves and updates information based on multiple conversations had over a period of time.