UNRESOLVED
Who Should Verify the Knowledge Base?
Attio’s support team is growing faster than any one person can keep its knowledge current. Cora Van Leeuwen built a weekly staleness scan in Claude Code to buy back time … but who’s actually responsible for verifying what’s true is still an open question.
Cora Van Leeuwen joined Attio, an agentic CRM, in January. Her first startup, her first CRM, and her first new team in five years. The product was shipping two or three changes a week. The knowledge base was growing just as fast. And nobody had quite caught up to the question of who’s responsible for keeping it right.
An honor system that outgrew itself
Before Cora took it on, Guru ran on the honor system: if you noticed something undocumented, you created the card, and you became its owner and verifier by default. That’s a reasonable way to build a knowledge base from zero. It worked fine while the team was small enough that everyone was close to everything.
It stopped scaling once Attio started shipping constantly. “It seriously felt like every single week we had two or three changes happening to the product,” Cora said of her first months. A few months into ramping up, she asked her manager if she could own Guru outright.
“That’s part of why we hired you.”
Building the staleness scan
Cora’s answer was a Claude Code skill she’s been iterating on all year: the staleness scan. It’s scoped on a fixed list of Slack channels and specific Guru collections, not all of Slack. Too much noise, too many tokens.
It looks for signals that something changed, matches them to the Guru card affected, decides whether that card belongs in Attio’s Fin-enabled or internal-only collection, checks its own prior runs so it doesn’t re-flag something already fixed, then drafts the edit as a Guru card — labeled “agent” in the title, so Cora always knows what came from Claude.
“I publish everything myself. It doesn’t auto-publish anything,” she said.
Ownership of verifying cards isn’t solved. Shortly before Cora started, Guru had been split evenly across the team — everyone got 50 to 100 cards to keep current. It was a sensible starting point. It stopped fitting as people specialized and the team grew, and it created a new wrinkle: whoever writes the most cards ends up verifying the most cards, whether or not they have the time.
“It’s not exactly feasible for one person to own all of the knowledge, unless that’s their only job,” Cora said.
The staleness scan solved detection of updates that fell through the cracks. It didn’t solve who’s responsible for verifying the fix. The team is consistently adding headcount but the product is scaling faster than the support team can keep up.
In the meantime, Cora’s interim answer is automation: not to replace the decision of who’s responsible, but to buy back the hours to make it possible for whoever that ends up being.
“How do we automate things, save time for ourselves, and keep from falling behind and drowning in everything that needs to get done?” she said. That’s still the unresolved part.
Links mentioned
Follow the show
Have a story to tell?
So many CX leaders are testing with AI and rebuilding with AI as their foundation. Every one of us have to reinvent everything that we’ve known about customer experience systems and processes.
You’re not alone, so let’s do this journey together.