How I AI
I use AI to write consistent UX copy
From
Copy written from memory, caught late in review.
To
First drafts already checked against the guidelines.
The challenges
How I used AI to solve it
Design-system guidelines and each project's shipped language live apart.
Packaged both together into one skill that drafts and checks copy against them.
Copy drifts between designers, screens, and releases.
Built on a deterministic classifier and an eval harness underneath it, not just a prompt.
Impact
Used weekly by designers and product analysts across the team to keep the system's many error and info-message copy consistent, without writing each one from scratch.
Across all three
What building my own tools with AI taught me
- 01
The output was only as good as what I fed it — curating real examples and guidelines was the actual design work, not the prompting.
- 02
AI was strongest on legibility: taking something buried in tickets, exports, or a spec and turning it into a thing a team could actually look at together.
- 03
Shipping a rough tool beat proposing one. What moved things forward was someone seeing it work, not reading about it.