Artificial intelligence news.
Straight from the original announcements.

Harvey's marketing team runs design projects end to end with agents in Figma

October 21, 2026Source: Harvey
Harvey's marketing team runs design projects end to end with agents in Figma
Image: Harvey

Harvey says its marketing team now lets agents run whole design projects from start to finish, using Figma's MCP server and newer frontier models. In a blog post by Shawn Farsai and Gabe Pereyra, the company describes a launch film made entirely by agents and a pipeline that produces tailored demo films for hundreds of sales accounts.

Key points
  • Harvey says agents now run whole design projects end to end, using Figma's MCP server and better frontier models.
  • The Harvey Artifacts launch film was made entirely by agents inside Figma, from storyboard to exported cuts.
  • A menu of 44 demo flows and an engine of about 750 lines of JavaScript produce tailored films for hundreds of accounts.
  • One batch of 36 films took 69 minutes, and overnight agents scripted and voiced the next 97.
  • Agentic Brand Guidelines, Product Map and Motion Primitives live as Figma libraries that agents reuse.

A launch film built inside Figma

According to Harvey, the launch film for Harvey Artifacts was made entirely with agents inside Figma. The agent built each scene as a Figma Motion frame and keyed every camera move, UI state and line of type on its timeline. It then exported the cuts straight from the file.

The process began with a storyboard, which the authors compare to a PRD in software: it sits between the creative brief and production, so every shot, line and transition is decided up front. The team reviews on the canvas, sometimes with 50 comments at a time. After each pass the agent reworks the board and asks which notes should become rules.

The agent also handled music. The team marked where the beat should drop and the agent retimed every scene to the track. When the ending fell flat, it cut and EQ'd the track, added sound design and proposed four or five ways the outro could land.

Tailored demos for hundreds of accounts

Harvey picked a sales problem to test scale: a tailored product film for each of the hundreds of accounts it wants to win. Early prompts gave results that were "half decent", and 15 to 20 minutes of back-and-forth shaped each one. That would not scale, so the team built a foundation in Figma in two parts.

  • Product Map: agents crawled the whole app, clicked every button and rebuilt all key screens as components.
  • Motion Primitives: a library of animated parts, from camera pushes and screen handoffs to a steady click rhythm, including product moves such as redlines arriving in a document.

Both libraries are published to Harvey's Figma org, so a fix to one component reaches every film that uses it. On top of them, the team built a menu of 44 company-agnostic demo flows in 13 categories, each built around one moment buyers ask to see.

For each account, agents use Enterpret's MCP to pull together Gong calls, Salesforce win-loss reports and Slack. An agent writes the script, shows a fictional company on screen, and voices it as the narrator Ashleigh Whittaker. Harvey built her voice in ElevenLabs from ten minutes of her footage.

Engine, parallel lanes and tracker

An engine of about 750 lines of JavaScript turns each script into a film inside Figma. Each film is a short list of beats, and the engine handles camera, transitions, captions, the pointer's click rhythm and the end card. It also builds contact sheets and flags clipped text, small type and stray pointers before anyone watches.

Scale brought practical problems. Only one agent can work in a Figma file at a time, so agents queuing in one shared file pushed calls to two minutes and caused timeouts. Harvey built a bus to run them in parallel: three lane files, each with its own builder agent, plus a coordinator for export, mixing and delivery. One batch of 36 films took 69 minutes. After the team hit its daily Figma call cap, overnight agents scripted and voiced the next 97.

Coordination runs through one tracker, a Claude artifact that every agent and person reads and writes. Each film has a row with its account, story, status and links.

Brand guidelines as a Figma library

Harvey treats its Figma like a codebase. Its Agentic Brand Guidelines act as the top-level AGENTS.md, and each project folder has its own AGENTS.md page that adds only what it needs. The company calls this progressive disclosure: an agent reads the brand, then the project, then only what its task requires.

Markdown files worked for design tokens such as colors and type, the authors say, but broke down for tasks like making a chart or structuring a long PDF guide. So the guidelines became a Figma library, from foundations up to templates with worked examples, so agents see relationships instead of reading about them.

Layout was where Figma beat HTML canvases, according to Harvey. Browsers set type in line boxes padded by the font's ascent and descent, so a headline's cap height never sits exactly on a grid line. Figma trims text to cap height and baseline, hangs punctuation and holds each grid as a layout agents apply.

Voice, images and a Slack-based creative agent

Every line of voiceover comes from ElevenLabs, which agents call through its MCP. Harvey's voice library holds clones of the narrator and of teammates, each built from a short clip. Figma Weave, a canvas for chaining image and video models that is now part of Figma, is reached through the same MCP. Agents run a model or a saved Weave workflow and place results directly in the file, guided by the art direction in the brand guidelines.

Spectre, Harvey's internal agent platform, is used by engineers to run coding agents and by the sales team as a data agent. Pointed at the brand repo and Figma files, it became a creative agent that works in a Slack channel. Someone tags it with a request, and it does the work in Figma, then replies with what it made, what it changed and which decisions to check.

Harvey argues this is what multiplayer design looks like now. People who own the work in product, legal, sales and marketing give direction in the Slack thread where they already work, while an agent does the production in Figma.

Related news

TalentNeuron and CodeSignal Expand Skills Intelligence Partnership
BusinessNovember 9, 2026, 14:47

TalentNeuron and CodeSignal Expand Skills Intelligence Partnership

TalentNeuron and CodeSignal have expanded their partnership to combine labor market and competitor insights with validated evidence of employees’ skills. The companies…

Gong Agent Builder connects revenue signals to sales action
Products and toolsNovember 9, 2026

Gong Agent Builder connects revenue signals to sales action

Gong has introduced Agent Builder, a tool that lets revenue teams build custom agents to turn sales signals into actions. The company says the offering combines…

Harvey Unveils Outlook Agents, iManage Sync and Fact Graph
Products and toolsNovember 5, 2026

Harvey Unveils Outlook Agents, iManage Sync and Fact Graph

Harvey has unveiled a set of product updates aimed at making its legal AI more personalized to teams and more specialized for practice areas. The announcements at Harvey…