AI enablementPublished v1.0

What 29 AI-assisted design tasks taught a team, written down as one playbook.

A design team's AI trial became a short, evidence-based guide: four plays, four rules and a clear list of what AI should not be asked to do.

29tagged production tasks reviewed
4plays, each with an exit point
The AI Design Playbook page listing which jobs AI does well and four working rules
Rebuilt document page · summary

The business problem

Designers were trying AI tools on real briefs, each in their own way. Results varied, prompts were lost with the chat, and the same mistakes repeated. The team needed a shared standard, built from what had actually worked.

What changed

Yash compiled every task tagged as AI-assisted between 12 July and 12 August 2026, along with the designers' own ratings and notes, into one playbook. It sets out four plays, a field-by-field way to turn a brief into a prompt, and a list of anti-patterns.

Each play has a trigger, a prompt scaffold and an exit point, the moment to stop prompting and go back to Figma. A short logging routine keeps the evidence growing.

Before & after

AreaBeforeAfter
Using AI on a briefTrial and error per designerFour named plays with exit points
Client contextRe-explained in every chatSet up once per client, then reused
Knowing when to stopSessions ran onA 12-minute timebox
Learning from resultsNotes scattered across tasksRatings and notes logged in a fixed format

The business value

The playbook turns a pilot into rules a designer can follow on day one. Its main findings: AI does three jobs reliably, it has not produced a finished on-brand ad, and sessions that ran to 15 minutes all scored 2 out of 5.

How to read the return

The evidence is 21 self-reported ratings from 7 designers over three weeks, with no control group. AI-assisted tasks had a 55% first-round approval rate, but the figure without AI is unknown. Nobody has yet shown that AI makes the team faster, and the playbook says so.

Why the guide is dependable

Every rule traces to a logged task or a designer's own words. Failures are documented as carefully as wins, and the six 2-out-of-5 sessions each produced an anti-pattern. The guide ends with a list of what is still unknown and is meant to be reviewed against approval rates at each pilot readout.

A closer look at the build

Four play cards with when to use each, an example note and an exit conditionThe four plays, each with an exit point (full size, opens in a new tab)

Pages are re-created from the playbook's structure. Quotes are shortened from the document.

Could this work for your team?

If your team is experimenting with AI tools without a shared method, a short evidence-based playbook can replace everyone's separate trial and error.

Talk through a similar project
Keep exploring

More case studies

All case studies