Shared AI content infrastructure
As a marketing consultant at Recce (a data validation tool for data and analytics engineers) longside Dori, their GTM lead, and the broader team, I helped shape what started as a content workflow into shared team infrastructure for AI-assisted work.
How it started
When the team began using AI for writing, everyone was pulling their own context into their own sessions. Five people, five slightly different setups, five slightly different outputs. The knowledge existed: voice guidelines, product positioning, terminology. It just wasn't shared anywhere that mattered.
My contribution was building the voice system: universal principles, content-specific voices for different post types, blog types, doc types. Not guidelines for humans to interpret but instructions precise enough for an AI to follow consistently. Once the team was working from the same source, the output started sounding like Recce without anyone having to review for tone.
How it grew
Dori's framing shaped how the whole system developed: a prompt fixes a problem once, context fixes it continuously. The team took that principle and applied it beyond content. Bug reports from Loom videos. Meeting notes sliced into project specs. Release notes after a deploy. All the same shape: repeating work that requires shared knowledge to do well.
Skills got built for each of these. Engineers started contributing their own. The system grew because the pattern was useful, not because anyone mandated it.
The architecture
Three layers: knowledge (what the team agrees on), workflow (how we do things), and context (where information lives). My work lived mostly in the knowledge layer: voice, content structure, terminology, product positioning. The workflow and context layers grew out from there, built by the team.
Packaged eventually as a Claude Code plugin called recce-team, with versioning and a shared install path so updates propagate to everyone automatically.
The numbers
- Voice system and content structure covering blog, social, and docs workflows
- 6 skills across content, operations, and engineering
- Adopted across GTM, content, and engineering teams
- Published as a Claude Code plugin
- Featured in: We Didn't Set Out to Build a Team AI Plugin
Why this one matters to me
The goal was always to make the knowledge transferable, not dependent on me being in the loop for every draft. The voice system working without my review is the thing I'm most proud of. It's also the clearest example of how I try to build: document it well enough that it runs without you.