---
type: "post"
slug: "figmatrace-dataset-for-figma-design-workflows"
title: "FigmaTrace dataset for Figma design workflows"
description: "Patronus published FigmaTrace, 3,469 Figma design trajectories covering 200+ hours of human work, on the argument that vision models fail at design for lack of workflow data."
retrieval_nugget: "Patronus published FigmaTrace, 3,469 Figma design trajectories covering 200+ hours of human work, on the argument that vision models fail at design for lack of workflow data."
published_at: "2026-09-02"
updated_at: "2026-09-12"
record_date: "2026-08-25"
date_kind: "discovered_at"
topics: ["agent-retrieval","agents","ai"]
entities: ["FigmaTrace","Patronus AI"]
source_url: "https://patronus.ai/blog/figmatrace-a-comprehensive-training-dataset-for-figma-design-workflows"
source_title: "Patronus AI"
source_domain: "patronus.ai"
source_terms: ["FigmaTrace","dataset","Figma","design","workflows"]
summary_word_count: 185
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:figmatrace-dataset-for-figma-design-workflows"
status: "published"
source_urls: ["https://patronus.ai/blog/figmatrace-a-comprehensive-training-dataset-for-figma-design-workflows"]
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:c5a2a051cc65a803e0c78a6813581636902a4fbc3ed5ec3292e1e3e916f011a3","reviewed_at":"2026-09-12T10:00:00Z","source_evidence_count":1,"verified_claim_count":2,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":2,"implication_count":1,"watch_condition_count":1,"related_record_count":2}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"Patronus introduced FigmaTrace, a dataset of 3,469 Figma design trajectories capturing more than 200 hours of human design work across ten creative skills.","observed_facts":[{"text":"FigmaTrace contains 3,469 Figma design trajectories covering more than 200 hours of human design work across ten creative skills.","source_urls":["https://patronus.ai/blog/figmatrace-a-comprehensive-training-dataset-for-figma-design-workflows"]},{"text":"Tasks include close-ended accessibility and templating fixes alongside open-ended prototype wiring, sketch-to-Figma and platform adaptation.","source_urls":["https://patronus.ai/blog/figmatrace-a-comprehensive-training-dataset-for-figma-design-workflows"]}],"mechanism":"Recording trajectories rather than finished files is the point: the artifact shows the order in which choices were made, which is the part a final Figma file destroys.","why_now":"Vision language models have progressed on objective, verifiable tasks — document understanding, robotics, computer use — while continuing to underperform on subjective creative design work, where design taste and figurative meaning matter.","implications":["Patronus attributes a major part of that gap to missing high-quality human workflow data that captures the preferences and decisions expert designers actually make."],"evidence_boundary":"This is a vendor announcement of the vendor's own dataset: it describes scale and task coverage without published baselines, licence terms or inter-rater agreement on the subjective tasks.","watch_conditions":["The condition to watch is whether models trained on FigmaTrace are evaluated by designers rather than by automated similarity to the recorded trajectory, because taste measured as trajectory-matching rewards imitation rather than judgement."],"related_records":[{"url":"https://newruntime.com/posts/proof-leaves-the-executor","relation":"Prior coverage of the record of work as the durable artifact."},{"url":"https://newruntime.com/topics/generative-ui","relation":"Topic hub for generated interface work."}]}
routes: {"html":"https://newruntime.com/posts/figmatrace-dataset-for-figma-design-workflows/","markdown":"https://newruntime.com/posts/figmatrace-dataset-for-figma-design-workflows.md","json":"https://newruntime.com/posts/figmatrace-dataset-for-figma-design-workflows.json"}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
---

# FigmaTrace dataset for Figma design workflows

## Retrieval answer

Patronus published FigmaTrace, 3,469 Figma design trajectories covering 200+ hours of human work, on the argument that vision models fail at design for lack of workflow data.

Patronus introduced FigmaTrace, a dataset of 3,469 Figma design trajectories capturing more than 200 hours of human design work across ten creative skills. The tasks are split between close-ended work such as accessibility fixes and design templating, and open-ended work such as prototype wiring, sketch-to-Figma conversion and platform adaptation.

The diagnosis behind it is specific. Vision language models have progressed on objective, verifiable tasks — document understanding, robotics, computer use — while continuing to underperform on subjective creative design work, where design taste and figurative meaning matter. Patronus attributes a major part of that gap to missing high-quality human workflow data that captures the preferences and decisions expert designers actually make. Recording trajectories rather than finished files is the point: the artifact shows the order in which choices were made, which is the part a final Figma file destroys.

Capturing the path instead of the output is the same instinct as treating the accepted artifact and its record as the durable asset, tracked in [when proof leaves the executor](https://newruntime.com/posts/proof-leaves-the-executor/). For interface generation specifically, the [generative UI hub](https://newruntime.com/topics/generative-ui/) is where this dataset's downstream models would show up.

This is a vendor announcement of the vendor's own dataset: it describes scale and task coverage without published baselines, licence terms or inter-rater agreement on the subjective tasks. The condition to watch is whether models trained on FigmaTrace are evaluated by designers rather than by automated similarity to the recorded trajectory, because taste measured as trajectory-matching rewards imitation rather than judgement.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Agent Retrieval](https://newruntime.com/topics/agent-retrieval.md): Explore the agent-retrieval topic hub.
- topic hub: [Agents](https://newruntime.com/topics/agents.md): Explore the agents topic hub.
- topic hub: [Ai](https://newruntime.com/topics/ai.md): Explore the ai topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)