Moda is an AI-native design platform for non-designers that uses a multi-agent system to generate and iterate on professional visuals. The architecture consists of three agents: a Design Agent (custom LangGraph loop), a Research Agent, and a Brand Kit Agent (both on Deep Agents). Key engineering decisions include a custom DSL instead of raw coordinate-based scene graphs, a triage-then-skills pattern with prompt caching, dynamic tool loading (12–15 core tools plus ~30 on-demand), and context scaling based on canvas size. LangSmith provides full observability for prompt iteration, cost tracking, cache hit analysis, and error diagnosis. The platform targets B2B teams needing polished, brand-accurate pitch decks quickly.

7m read timeFrom blog.langchain.com
Post cover image
Table of contents
The Challenge: Making AI Good at Visual DesignAgent Harness: Built on Deep AgentsContext Engineering: The Details That MatterUX: The Cursor Moment for DesignObservability with LangSmithResults & What's Next
542 Impressions