Stacks/Design / content
LangChain's Content Studio design agent
LangChain's Content Studio is a Slack design agent on Managed Deep Agents with a sandbox per thread, async subagents, skills, a brand and motion wiki, code gates, and two-layer file memory. It has delivered over 3,000 assets, three in four requests now come from outside marketing, and the median request costs about 50 cents in model spend.
Amal Irgashev and Shreya TalegaonkarPublished Oct 27, 2026Last verified Nov 13, 2026
- Use case
- An in-house design agent in Slack that turns a brief into on-brand videos, social graphics, recreated UI images, and slide decks, all written in code
- Category
- Design / content
- Team size
- Built by LangChain's GTM Engineering team with the design team
- Orchestrator / framework
- Each Slack thread gets its own sandbox that keeps the project between turns. A main agent writes subagent briefs (role, context, file ownership, validation) and picks model and effort per task; long jobs run for hours on async subagents that coordinate through the project folder. The thread shows the plan, live progress, and an approval before anything ships, and people can steer mid-run.
- Memory
- Plain files in two layers, one shared across the team and one per person, with a fast loop that captures feedback as it's given and a slow loop that learns from completed projects. The always-loaded file is capped at a 200-line index after an early version grew to 500K tokens.
- Evals
- Measurable checks are 'gates', deterministic checks written in code. The agent also builds its own frame-by-frame checks on renders.
- Observability
- not stated
- Cost (stated in writeup)
- Median request costs about 50 cents in model spend (down from about $8 in mid-September) · per task
- Cost per output (stated)
- Stated: ~50¢ per request (median, model spend)Down from about $8 in mid-September
“The median request now costs about 50 cents in model spend, down from about $8 in mid-September.”
As stated in the writeup, checked 2026-11-13. Separate from the WhatWeRun estimate below.
- WhatWeRun estimate
WhatWeRun estimate
Not enough info to estimate
The writeup states the cost per request (shown above), so there's nothing to estimate. The stated cost stands.
Reviewed Nov 13, 2026
- Models
- Observability tools
- not stated
- Other tools
- LangSmith SandboxHyperFrames (HeyGen's open-source video framework)Paper canvas via MCP (Paper Desktop run headlessly)Skills and a brand/motion wikiMCP tool searchSlack
- Source
- https://www.langchain.com/blog/design-agent
Amal Irgashev and Shreya Talegaonkar · Published Oct 27, 2026 · Last verified Nov 13, 2026
Related
