Challenge
Research-to-prototype loops usually break at the handoff. A researcher learns something valuable, then waits on a designer or engineer to make it real, and by the time it's built the urgency has faded. Generic AI tools don't fix this, because they aren't grounded in how a specific team actually works — its strategy, its design system, its taste. I wanted researchers owning more of the prototyping themselves, with AI support that was specific to Savvy rather than generic, and with guardrails so nothing sensitive auto-published to the whole org.
Solution
I built the workflow as a chain of focused skills — each does one job and hands off to the next — grounded in shared context and gated by human review.
A chain of skills that hand off to each other
Four skills run the loop: research prep (strategy-grounded planning), synthesis (findings pulled from transcripts, with citations required), design provocation (distinct design directions), and design crit (structured prototype critique). Each does one thing well and hands to the next. You don't need all four to get value — research prep plus synthesis alone is already a meaningful upgrade.
Grounded in shared context, not generic
The skills read from canonical context documents — the team's strategy, design system, and design principles — that live in one place. So every research plan is cross-checked against strategy before drafting, and misalignment surfaces before someone has written five pages. One source of truth, no drift between skills.
Humans control what publishes
Nothing goes org-wide without explicit human approval. The skills stay conversational — refining in chat until the person signs off — and treat the final doc or page as the artifact, not the working surface. There are two review gates before anything publishes to a team-wide channel, because auto-publishing is high blast-radius.
Built to be forked
The Savvy version references our internal tools — Notion databases, Slack channels, specific Drive folders, the Savvy Kit design system. I stripped those specifics out and turned the chain into open-source templates with clearly marked placeholders, so any team can drop them into their own marketplace and wire them to their own tools.
Impact
The chain is in active use by the Savvy design team — researchers run the research-to-prototype loop themselves instead of queuing behind design and engineering. Extracting it into open templates means the underlying pattern is reusable by other orgs, not locked inside our stack. The point was never the skills themselves. It was researchers owning the prototyping work; the skills are the support system that makes that possible.
Reflections
Building these clarified for me which design choices make an AI skill useful instead of decorative: citations are non-negotiable, context lives in one canonical place, and humans hold the publish button. Those aren't accidents — they're the difference between a tool people trust and a demo.
It's also the clearest example of where I think design leadership is heading. The highest-leverage thing I can build isn't a screen. It's the scaffolding that lets more people do the creative work directly.