What's next for agent AI? The founder of LangChain looks to environmental representatives

Agent AI is the latest big move in generative AI, but what comes next?

Although it seems that full artificial general intelligence (AGI) is still some time in the future, there may be an intermediate step with an approach called environmental agents.

LangChainthe AI ​​agent pioneer, introducing the term “environmental agents” on January 14. The technology that LangChain develops includes the unsigned open source LangChain framework that enables organizations to connect different large language models (LLMs) together to obtain a product. LangChain Inc. raised $24 million in funding in February 2024. The company also has a series of commercial products, including LangSmith for LLM Ops.

With traditional AI interfaces, users typically interact with LLM through text prompts to initiate an action. Agent AI usually refers to LLM-powered systems that perform actions on behalf of the user. The concept of environmental agents takes that paradigm a step further.

What are environmental agents?

Environmental agents are AI systems that run in the background, continuously monitor event streams and then are prompted to act when appropriate, based on pre-instructed instructions. fixed and user intent.

Although the term “environmental agents” is new, the concept is environmental awarenesswhere AI is always listening, no. Amazon refers to its Alexa personal assistant technology as enabling environmental intelligence.

The goal of environmental agents is to automate repetitive tasks and scale user capabilities by running multiple agents continuously, rather than having a human user invoke and interact with each one. , one-to-one. This allows the user to focus on high-level tasks while the agents handle routine work.

To help validate and advance the concept of environmental agents, LangChain has developed a series of original use cases, one that monitors emails, the other for social media, to help users manage and respond when needed.

“I think producers in general are powerful, inspiring and cool,” Harrison Chase, co-founder and CEO of LangChain, told VentureBeat. “Environmental agents are much more powerful if you have a bunch of them doing things in the background, you can just do more for yourself.”

The technology uses many open source solutions, and LangChain has not yet indicated how much it would cost to deploy any new tools.

How environmental agents work to improve the usability of AI

As with many great technological innovations, the original motivation for environmental agents was not to create a new paradigm, but to solve a real problem.

For Chase, the problem is one all too familiar to many of us: an overloaded email inbox. Chase began his journey to create environmental agents to solve email challenges. Six months ago he started building an environmental agent for his own email.

Chase explained that the email assistant categorizes his emails, handling the triage process automatically. He no longer has to manually configure it through his inbox, because the agent takes care of it. Through using the agent inbox over a long period of time, Chase was able to refine and improve his abilities. He noted that he started imperfectly, but by using it regularly and dealing with the pain points, he was able to increase the agent's performance.

To be clear, the email helper is not some kind of simple rule-based system for classifying email. It is a system that understands his email and helps him decide how to manage it.

The environmental agent architecture for the email assistant use case

The architecture of the Chase email helper is quite complex, involving several components and language modules.

“It starts with a triage step that's a bit like LLM and a very complex stimulus and a few short examples that are semantically retrieved from a vector database,” Chase explained. “Then, if he's determined he should try to fit in, he goes to a draft agent.”

Chase also explained that the agent has access to additional tools, including a sub-agent specifically for interacting with the calendar:

“There's an agent that I have specifically to interact with the calendar, because LLMs are very attractive on dates,” Chase said. “So I had to have a special agent just to interact with the calendar.”

After the draft response is created, Chase said there is an additional LLM call that rewrites the response to ensure the tone and format are correct.

“I found it difficult for an LLM to try to call all these devices and pick up an email and then write in the right tone, so I have a degree specifically for tone,” Chase said. .

The agent inbox as a way to control agents

A key part of the environmental agent experience is having control and visibility into what the agents are doing.

Chase noted that in the first implementation, he only had agents message through Slack, but that quickly became inconvenient.

Instead, LangChain designed a new user interface, the agent inbox, specifically for interacting with environmental agents.

The system displays all open lines of communication between users and agents and makes it easy to track specific activities.

How to build an environmental agent

LangChain is primarily a tool for developers and will become a tool to help build and deploy environment agents now as well.

Any developer can use the LangChain open source technology to build an environmental agent, although additional tools can simplify the process. Chase explained that the agent inbox he built is somehow a view on top of the LangGraph platform. LangGraph is an open source framework for build agents that provides an infrastructure for running long-running background jobs.

In addition, LangChain uses its commercial LangSmith platform, which provides visibility and evaluation to producers. This will help developers put agents into production with the necessary monitoring and evaluation tools to ensure they are performing as expected.

Environmental proxies: A step towards the use of general information

Chase is optimistic that the concept of environmental agents will catch on with developers in the coming months and years.

Environmental agents are expected to give AI even more autonomy, allowing it to monitor the flow of events and take intelligent actions. Chase still expects the need to keep people in the loop as part of the environmental agent experience. But people only need to confirm and confirm actions, rather than finding out what needs to be done.

“I think it's a step toward using and using information more broadly,” Chase said.

Chase noted that true AGI is more likely to come from improvements in reasoning models. That said, doing better practice of models where the concept of environmental agents will give value.

“There's still a lot of work to be done to make use of the models, even after they become very intelligent,” Chase said. “I think the environmental agent interface style will absolutely be the solution for the to use this general information”.

An open source version of the email assistant is currently available. LangChain is releasing a new social media environment agent today, and will make an open source version of the agent's inbox available on Thursday, January 16.