LangChain has open-sourced Content Studio, an agent that creates videos, graphics, slide decks and interface mockups from requests posted in Slack. Each project receives a persistent sandbox containing code-based design tools, a shared canvas and the files needed to inspect and revise its own output.
The company says employees have used the system to produce more than 3,000 assets, including over 80 decks. Three quarters of requests now come from outside marketing, and 75 percent of users return weekly. LangChain reports that median model cost fell from about $8 to 50 cents per request after migrating the system to its Managed Deep Agents platform.
The project’s main lesson is that a capable model still needs organization-specific context. Designers documented brand tokens, visual principles and reasons behind feedback rather than relying only on rigid rules. LangChain also removed workarounds as models improved, keeping durable design principles while discarding constraints built around old model weaknesses.
Long jobs can run for hours through asynchronous subagents, so the Slack thread shows a plan and live progress and requires approval before work ships. Those controls matter because polished output can still be off-brand or communicate the wrong idea. The repository contains the full harness, but organizations adopting it must supply their own brand guidance, review process and deployment environment.