AI Workflow
101
Agenda
Why now:
agents on your desktop
Three reasons
The shift is
already at your desk
Your files are the interface
Local agents read your actual folders, Figma exports, and code — no upload, no re-typing the brief into a chat window.
Off-the-shelf agents are already strong
Claude Code, Codex, and Cursor already sit on employee laptops. The workflow needs wiring, not a new model to train.
Every session compounds
Preferences and design systems are saved as memory, so the tenth request needs less explaining than the first.
Steps, not a black box
Every request runs the same loop: read the brief, propose a change, execute it locally, and log the result for review — repeatable enough to put in a training deck.
We didn't need a new tool budget — our own coding agent already knew how to open a file. Open Design just gave it the workflow.
— Pilot lead, 12-person design team
Unmanaged pilot vs. Open Design guardrails
Unmanaged AI pilot
- No file audit trail, no version history
- Any employee, any model, no key controls
- Output lives in someone's chat history
- Risk review happens after launch
Open Design guardrails
- Git-diffable changes, fully version controlled
- BYOK: your own model keys, your own usage caps
- 21 supported coding agents, chosen by IT
- Local-first: nothing leaves the laptop by default
The workflow map
Read
Agent opens the actual project folder — design files, code, docs — no manual export step.
Propose
Drafts a concrete change: a new page, an updated component, a slide deck.
Execute
Runs the change locally and shows a diff, exactly like a pull request.
Verify
A named reviewer approves or reverts before anything ships.
Pilot data, 6 weeks