Rethinking AI Agents: Why a Simple Router May Be All You Need
AI Agents have been hot topic first started with the hype around AutoGPT and BabyAGI. From these projects, a wave of new tools emerged promising to simplify how we build applications with Large Language Models (LLMs). By far the most common one I hear about is LangChain and specifically LangChain agents.
While LangChain agents offer great out of the box features, in many use cases you don’t actually need an agent. You just need a router that decides which part of your code to call. In essence, this becomes a simple classification problem, not an agent problem.
In this blog, I’ll provide a background on how agents work, some side effects of using agents, and define a simple software pattern that can used in lieu of agents for many common use cases.
The idea behind agents is the use of an LLM as a reasoning engine to decide which actions to take so that a task can be completed. An agent uses tools to augment it’s capabilities in order to complete these tasks. As an example, if the agent needs to google something, you could provide a search engine (SERP) tool to call Google and return the results back to the agent.
