LangChain Opens Content Studio, the Slack Design Agent Behind 3,000 Assets
LangChain open-sourced Content Studio, the Slack-based design agent that has shipped 3,000+ on-brand assets, built on Deep Agents.
Takeaways
- LangChain open-sourced Content Studio, its internal Slack design agent, under Apache 2.0.
- Built on Deep Agents and Managed Deep Agents, using Claude Opus 5.5 by default.
- Produced 3,000+ on-brand assets internally; median request cost dropped from $8 to about 50 cents.
- Each Slack thread gets its own sandbox with Chrome, FFmpeg, HyperFrames and brand fonts.
- Every asset is code (HTML slides, animation timelines) with automated brand and chart-math checks.
- Deploys with three CLI commands via Managed Deep Agents, or self-host with Docker.
LangChain, which develops tooling for AI agents, has released the internal system it uses to create marketing graphics, slide decks, charts and launch films. Content Studio works inside Slack: a user mentions the agent in a thread, supplies a brief and reference files, and receives finished, on-brand assets. The repository uses the Apache 2.0 license, and LangChain’s project write-up describes it as the system running in production.
The repository exposes a practical architecture for agents that produce editable files through long-running, multi-step workflows. Developers can inspect the skill router, sandbox configuration, brand controls, memory system, subagent coordination and deployment path rather than infer them from a product demo.
| Metric | Reported result |
|---|---|
| Assets delivered | More than 3,000 |
| Median model spend | About $0.50 per request, down from about $8 in mid-September |
| Internal agent code | About 80% smaller after moving to Managed Deep Agents |
| Reach on X | About 4.5 million views on posts featuring its work |
| Requests outside marketing | About 75% |
LangChain reports these figures from its own deployment. The 50-cent median covers model usage only; hosting, infrastructure and engineering costs fall outside that number.
The repo stops at the design layer
Content Studio combines Deep Agents with Managed Deep Agents, or MDA, and uses Claude Opus 5.5 by default. Deep Agents supplies the planning loop, file operations and subagent model. MDA provides the managed Slack app, isolated sandboxes, memory, schedules and tracing.