Cadence
An AI copilot that designs a feature's entire launch
A product-communications system for a design platform
How a company turns one launch into a full, on-brand rollout without five teams building five versions of it.
I defined the problem, designed the operating model, the agents, who owns them, and the rules they run on and built the working prototype you can try below, from the input brief to the governed review. Shown on Figma as the example for context.
Try it, in two steps
The system is easiest to understand by using it. First you give it a brief. Then you watch it get reviewed and shipped . the same rollout, start to finish.
Step 1 • The brief
You don't write a prompt and hope. You make five choices, in plain language, and run it once. Change any field and run it again.
The design team asks for a rollout.
A constrained brief, not a chatbot. The team states what needs to go out. Cadence writes a version for every discipline — you'll review them all in the next step.
One designer reviews every version.
Same rollout, a voice for each discipline. The reviewing designer checks all four — one message per audience — then submits it for internal approval.
The second approval, then testing.
Governance checks the whole rollout against the system, the brand, and an attention budget in one pass — then it's pushed into testing to await validation before a live publish.
Pushed into testing
The rollout is on the testing branch, waiting for validation. Nothing reaches users until it passes — then it's cleared for a live publish to the master branch.
Step 1
Handed the system a brief, and it composed the whole rollout.
Step 2
Where it gets governed and where it decides what's actually worth sending.
Step 2 • The review
Every design is checked against the system, the brand, and an attention budget in one pass. Most of it clears itself; the rest routes to the team that owns the call.