Deep Agents (LangChain)

What was covered

• What it is. DeepAgents is built on LangChain Core and lets you build agents with more advanced capabilities.
• What you get. An execution environment with virtual file systems, context management, delegation to sub-agents, long-term memory and a human-in-the-loop mechanism for supervising a run and approving changes to files or commands.
• A demo. An orchestrator agent used search tools to collect information on sports matches.

What came up

• When to use it. Asked about ideal use cases compared with a standard create_react agent, Pedro said DeepAgents helps with file manipulation and complex execution. The team will probably keep using create_react agents in 99% of cases, because they are simpler and have lower latency.

Also on the agenda

• This site. Felipe and Bruno are consolidating the guild’s information repositories into one website, to formalise the stack, share best practices, onboard new members and show other guilds what the team does.
• Backstage. Eduardo suggested Spotify’s developer portal to map repositories and scaffold new ones, for example AI backends. Felipe agreed to look at how it could work with the existing Markdown documentation.
• Scaffolding. An internal effort to create project templates exists, but progress is slow because people are on full-time projects.