Voice Velma

When I joined Resident, none of the four DTC sleep brands I wrote for had a defined voice — every team was guessing at tone independently, and anything that needed real brand judgment had to route through me directly, one request at a time.

The system started small: I trained my own Claude on each brand's voice, just to speed up my own writing. As it proved useful, it became a shared resource — separate Claude Projects for each brand, so any team member could reference the right voice without pulling me in. Over the next year, as our brand roster grew to six (including Newton Baby, acquired during this period, and a sixth brand not yet launched), that per-brand structure started to strain. The real shift came when we moved the entire reference library into Notion alongside the rest of our content ecosystem. This live structure is what made it possible to build one joint agent instead of six separate tools.

This wasn't a case of building an agent to fit our existing systems. Instead, we restructured our infrastructure to fit her. Velma consults a Notion reference library at the moment she's writing: brand voice guides (which always win), shared format rules by content type, and brand-specific examples where they exist. Product facts — specs, dimensions, certifications, availability — come from a live knowledge base she reads in real time, so she's never inventing a detail that isn't true. She's triggered by an @-mention or within Notion's agent dashboard, and, most crucially, I can see her chats with team members to run consistent quality control.

Velma doesn't simply provide copy. She writes PDPs that populate directly into our PDP library and blog posts that populate directly into the blog database. There's no copy-paste handoff between "what she wrote" and "what's live in the system."

Six brands, one system, and zero hallucinated specs. My personal productivity is up 85%, and that productivity is now being passed on organization-wide.