The brief, every quarter: faster design, same headcount, AI as the lever.
Asked for speed, delivered it with AI, team of four
Handoffs per designer more than doubled. A 5-hour Figma rebuild became one click. The time that freed started the company's first research practice.
Measured July to October 2026.
- Team
- 4 designers, with an org-wide AI specialist
- Product
- QuintAce, an AI poker coach on WPT Global
- Scope
- All design squads
Design's share of the product's AI-tooling commits rose from 42% to 88%
Situation
Engineering had agents. Design had a chat box.
When I joined, engineering had coding agents and its own tooling. Design had ChatGPT, no repo access, and every prototype was rebuilt in Figma by hand. The ask was to close that gap and get faster, with four designers.
Leverage first, then let designers build for their own pain
- March Git and Claude Code for every designer. Weekly sessions on branches and pull requests. First commit 20 March. Four designers committing in four weeks.
- April to June One pain point, one command: bind-tokens, design-qa, find-figma. 9,658 token bindings, zero errors, in a first pull request. A brand re-theme in one sprint.
- July to August A skill for each phase, discovery to ship. A Slack agent for component requests. An Education Hub. 37 tools, three languages. 22 of 35 requests fixed in six weeks.
- September to October Prototype-first: the prototype goes to engineering, and a bridge drops it into Figma as library components. 5 h → 1 click. 5 tickets built straight from a prototype.
Two decisions that cost something
Skills first, the agent last
I built an orchestrator agent in March without asking designers what hurt. I shelved it and inverted the order.
Cost: our most demo-able thing.
Prototype-first delivery
The prototype is what engineering builds from. Figma polish runs in parallel.
Cost: Figma is no longer the only source at handoff.
The week runs itself
Definition of Ready and Done. Jira opens design and QA tickets on its own. A management agent runs my week:
- MondaySprint commitments to each designer; cross-team priorities.
- TuesdayAction points from the product meetings.
- ThursdayBacklog planning sync.
- FridaySprint close, with stakeholder updates.
Less Figma, more players
Moved to AI
Variations, hi-fi tweaks, handoff prep, Jira writing, docs.
Stayed human
Problem framing, deciding, peer review, talking to players.
Research, from zero. With the capacity freed, the first moderated study ran: five think-alouds that showed players couldn't reach their analysis unassisted. The fixes to onboarding and analysis shipped. Synthetic personas built from live player data now give a first read before each handoff.
| Metric | Before | After |
|---|---|---|
| Hours to rebuild a prototype in Figma | 5 | One click |
| Design tickets with a clickable prototype | 9% | 50% |
| Handoffs per designer, per week | 0.8 | 1.8 |
| Share of AI-tooling commits from design | 42% (Mar) | 88% (Sep), 77% overall |
Sources: Jira, Slack, Figma version history and Git, March to October 2026.
Learned
Adoption is the work
The team built 37 commands. Weekly sessions got the first commit; a few hands made the volume; now every designer ships from prototypes. Ask designers what hurts before building for them.
Next
- Hours saved: tracking is live; first read next quarter.
- A product metric: the first retention read on the redesigned onboarding and analysis flows lands next quarter.
- A buddy developer per designer, so handoff quality improves as much as handoff speed did.