LangChain recently introduced Deep Agents: a new way to build structured, multi-agent systems that can plan, delegate, and reason across multiple s...
For further actions, you may consider blocking this person and/or reporting abuse
Thanks for this walkthrough! The step-by-step breakdown of building a frontend for LangChain Deep Agents with CopilotKit was super clear and practical. I especially appreciated the way you tied UI components to the agent’s logic — that makes it much easier to follow. Great job! 👏💡
Great Job Anmol!
Thanks Uli. I'm sure people will build crazy stuff using Deep Agents :)
Well written article, Anmal!
Thanks Bonnie! UI is a bit messy lol (most of the time went in improving system prompt) but hopefully others build something cool with this pattern.
Good to see you back creating great articles.
This is the part of agent dev people keep skipping: the frontend. Building the agent is the easy flex — getting state, tools, and outputs to flow cleanly into a UI without dumping JSON into chat is where it gets real.
I like how you framed Deep Agents as “plan → delegate → externalize context” instead of the usual “LLM in a loop and pray.” And the CopilotKit + AG-UI sync makes a lot of sense because the agent’s state is already explicit (todos, files, messages). That’s the missing bridge.
Also: the strict prompt rule of “jobs only via update_jobs_list” is such a simple move but it solves a huge UX problem. Keeping chat conversational while structured data renders separately is the difference between a demo and a product.
Good write-up — this actually makes Deep Agents feel buildable, not just blog-hype.
yeah, I went through like 10+ different system prompts (it got a bit crazy 😅) and this was the one that finally worked after a lot of refinement. In the end, Deep Agents have a really powerful architecture pattern & using CopilotKit gives you a lot more visibility into what’s actually happening.
Great walkthrough, Anmol!
I've been waiting for a tutorial on Deep Agents, this is great!
Thanks Nathan. the concepts of subagents is really cool and the entire architecture of deep agents is actually impressive -- I'm definitely going to go really deep in their docs and try more stuff.
Nice writeup!
This really helped my understanding of how and when I can use a Deep Agent
Love the detail here
This is awesome Anmol! I know a lot of people who have been struggling to build frontend capabilities for their deep agents. This makes it way easier
yeah deep agents provides some really cool stuff built-in and I learned a lot while building this.
Very helpful. Great detail and I have a better understanding of which pieces fit where. Thanks
Well explained