LangChain open-sources its Slack design agent, with brand judgment built in

LangChain has open-sourced Content Studio, the design agent it built to turn Slack requests into videos, graphics and slide decks. In its October 27 release account, the company describes four months of teaching the system its design judgment—a process that went beyond brand colors and fonts to explaining why a design works.

First, teach it what the brief leaves out

The starting problem was a stretched design team. Product campaigns, webinars, blogs and events all needed assets, and repetitive requests were crowding out more strategic work. LangChain built Content Studio so colleagues could describe an assignment in Slack, attach a brief or supporting material, and receive finished files.

Early results fell short. The model could write layout and animation code, but it lacked the company’s context. In one comparison, a local coding agent given brand instructions still missed the correct logo arrangement, text alignment and distinction between primary and secondary colors. LangChain says Content Studio handled the same prompt correctly.

Its response was to record more than visual specifications. Designers added principles such as one focal point per frame, warnings against common AI-generated patterns, and the reasoning behind their feedback. Automated checks could catch capitalization mistakes; subjective decisions about a campaign’s idea or message required designers to explain their judgment in reusable instructions.

Give each thread a working computer

The next layer was the working environment. Each Slack thread gets an isolated Linux computer through LangSmith Sandbox, with project files retained for the thread’s lifetime. Briefs, plans, scene code, audio and rendered outputs share one project folder, which the main agent and its helper agents work from.

Every asset is built in code. A slide can be an HTML page; a video can be an animation timeline. That lets the agent revise a timing value and render again without rebuilding unrelated artwork or copy. LangChain contrasts this with a flat image or generated clip, where a small edit may require recreating the asset.

For video editing, Content Studio uses HeyGen’s open-source HyperFrames framework, which represents video elements as a web page. Original animation uses LangChain’s own motion-studio engine. Longer jobs run through background helper agents, while Slack displays plans and progress. Teammates can steer work mid-run, and an approval step comes before anything ships.

Then remove the rules that stopped helping

More instructions were not always better. The first version included 115 reusable components for titles, layouts and animations. LangChain says this helped maintain its brand but encouraged repetitive compositions. It attributes a later quality jump to Opus 5.5, after which it removed most hard-coded rules while keeping design principles and high-level guidance. Variety improved further, the company says.

Content Studio has now delivered more than 3,000 assets, including over 80 slide decks, according to LangChain. Three in four requests come from outside marketing, and 75% of users return weekly. The company describes designers’ role as moving toward creative direction: setting the direction and reviewing work rather than producing every asset themselves.

The released repository includes the surrounding agent software, sandbox setup, instructions, memory and Slack experience. LangChain directs teams to fork it, add their own brand tokens, fonts and instructions, and deploy through Managed Deep Agents. Its own experience leaves a clear maintenance task: retain the design judgment, but revisit rules written to compensate for an older model’s weaknesses.