written by Matteo Castiello https://www.linkedin.com/in/matteocastiello/
Website: https://insurgence.ai/
Agents are all the rage however let's make one thing clear:
They are not a solution to a business problem.
They are a solution to a technical problem.
The only reason that someone would create an 'agent' is if they reach a limit on what more simple approaches for GenAI stop working e.g. basic RAG.
This seems to be the biggest misconception of an agent´s role...it's a technical problem solution...not a business problem solution.
Let's take a step back and define an agent:
An agent is a self organising system that interacts with it’s digital environment to reach a desired outcome.
That's a few steps further from a typical conversation with ChatGPT where the organisation/orchestration is dictated by a human.
This page describes some of the key components of an agent:
1. Memory: Agents can have access to remember past work in order to inform future tasks.
2. Tool use: Agents can have access to tools that allow them to communicate with external apps and data sources.
3. Planning: Agents create, execute and iterate on a multi-step plan to achieve a goal.
4. Action: Agents can interact autonomously with their digital environment, including interacting with other agents to perform tasks together.
Imagine that there are many agents composed of different components (tools, planning, memory, etc.) and they are working together to complete a task.
Very useful, but overkill for many of the low hanging fruit use cases for generative AI.
Where do you see agents being of the highest potential?
Full credit to this diagram goes to Harrison Chase from Langchain: https://lnkd.in/deNSDnCw
Comments