Automate agent-readable, shared knowledge systems
I partnered with teams that owned Google-wide agents, and built all tooling upon that standard.
TradeoffWith vibe coding, it seemed instinctual to create standalone sites, but I intentionally chose something agents could update and read.
While I could have gotten more UXD resourcing, my investigations with eng hit constraints where a visual gallery app for AI resources could not be kept fresh. We did not have resourcing to manually maintain it. I doubled down on the agent-readable solution and built out the infra.
How I built intel to be delivered at rapid pace
A system diagram. Internal chat channels, scheduled Gemini deep-research agents, industry reports, expert writing, newsletters, talks, job listings, code repositories, community sentiment and internal research all feed my GChat agent, which connects through MCPs to NotebookLM, Workspace and Docs. From there the same knowledge base completes three journeys, each ending in a surface people open: querying it in NotebookLM or an internal AI assistant; reading it as a formatted doc that Apps Script pushes automatically; or receiving it by email, which hundreds subscribed to.
What changed in the design process?
Before
After
Days of manual digging
Prior research sits across old docs and decks, and every study has to be moderated live by a researcher.
Agents find the source material
Research agents search existing docs and can moderate sessions. Deep Research adds current external sources and citations.
Research agentsGemini Deep ResearchTurn the first idea into a deck
People spend days building the deck before anyone challenges the idea, and nothing in it is interactive.
Run a critique pass
My metaprompter writes expert prompts. Designers launch multiple adversarial agents to critique plans.
Xinni's MetaprompterGemini Deep ThinkStatic, pixel-perfect mocks
The interaction still only exists in the mock.
Build a working prototype
Designers prototype in code using the agent skills from my tutorial.
Xinni's /Skills TutorialXinni's Automated Resource LibraryEngineering rebuilds it
Every interaction and edge case has to be translated into production code.
Designers push code
Designers push working changes with code scaffolding and stay close to engineering.
Code scaffoldingImpactMy data pipeline fully automates daily industry shifts and resource galleries, enabling me to create the agents that UXers use to adopt the latest workflows and tools. I regularly serve as the go-to for intel for Leadership's various AI transformation programs.
