How FLORA shipped a creative agent on Vercel's AI stack
By Steven Van ·
FLORA replaced LangChain and Temporal with Vercel's AI SDK, Workflow SDK and Fluid compute to run its FAUNA creative agent.
FLORA, the design tool built around a node-based creative canvas, has released FAUNA, an agent that turns a written brief into campaign visuals, moodboards or lookbook directions instead of requiring a manually configured workflow. To build it, FLORA's team moved off LangChain and Temporal and onto Vercel's AI stack.
The earlier canvas required creatives to configure each step and model choice by hand, interrupting the creative process. FAUNA removes that setup: users describe what they want to create, and the agent pulls references, chooses models and generates variations to explore, fanning out into many concurrent generation jobs across different angles, backgrounds and styling directions.
- AI SDK's agent framework supplies the models, tool calling and interface layer behind FAUNA's behaviour.
- Workflow SDK's DurableAgent adds durable orchestration, so each step in the loop persists and retries on failure without losing state.
- Fluid compute runs the parallel, long-running image generation jobs cost-efficiently.
FLORA says the move made its generation system 2x faster to reach production, ended infrastructure debates on the team, and lets it orchestrate more than 50 image models.
Read more in Vercel's case study.
