Skills Libraries and Agent Memory: Notion's Big Infrastructure Week

Notion shipped the infrastructure layer for agentic work (shared skills libraries, persistent agent memory) in the same week Teresa Torres published a deep case study on Vistaly rebuilding its core workflow around AI-assisted interview synthesis, and LangChain quietly dropped a 64% cost reduction on coding tasks just by routing models smarter, three signals pointing at the same thing: the agent tooling conversation is moving from "can it do the task" to "how do we govern, remember, and afford doing the task at scale." ### Notion - A Skills Library for Every Agent **Source:** https://www.notion.com/blog/a-skills-library-for-every-agent **Credibility:** High (first-party product announcement) **What happened:** Notion launched a skills library that lets teams manage agent skills using Notion's existing editing, collaboration, and permissions model, then deploy those skills across any agent that connects to Notion. Geoffrey Litt (tracked person, now at Notion) is credited on the post. It ships alongside a companion feature, "Lore," which turns agent conversations, decisions, and tasks into durable Notion pages that future agents and humans can both reference. **Key technical details:** - Skills are authored and versioned like any other Notion page, with the same commenting, permissions, and history - Skills library is designed to be used "everywhere" an agent connects, not just inside Notion's own AI features - Lore (the shared memory layer) persists conversations, decisions, and "durable facts" as structured Notion pages rather than ephemeral chat logs - Both features build on Notion's existing Custom Agents and MCP connections, which the company separately detailed in a case study about how its own Customer Education team uses agents to keep Academy content current **Why it matters for PMs:** This is the team-infrastructure problem for agents, the same one Wispr Flow solved for voice tools a few weeks back. An individual can prompt an agent effectively, but a team can't scale that without a shared place to store "what good looks like." Notion turning its own document model into the skills and memory substrate for agents is a strong bet that governance and knowledge-sharing, not raw capability, is the next bottleneck for agentic adoption inside companies. **Critical questions:** - Does a skills library actually get maintained, or does it rot the way wikis and runbooks always have? - How does Lore handle conflicting or outdated "durable facts" as decisions change over time? - Is this a moat for Notion, or does every agent platform need to build an equivalent and this just becomes commodity infrastructure? **Action you could take today:** If your team is already using Custom Agents or MCP connections into Notion, audit what "tribal knowledge" currently lives only in Slack threads or people's heads, and see if it's a candidate for a first skill or Lore entry.