Notion has officially embraced AI agents as a core platform feature with the launch of Custom Agents, marking a significant evolution for the productivity giant that has been experimenting with AI tooling since before ChatGPT. The feature, teased at Notion's recent Make conference, represents the culmination of years of iterative development and multiple complete rebuilds as the company grappled with fundamental challenges in early agent implementations. Sarah Sachs and Simon Last, key leaders in Notion's AI efforts, revealed the technical and organizational decisions behind the ambitious initiative. Early attempts at agent integration failed due to the absence of tool-calling standards, insufficient context windows, unreliable model performance, and excessive complexity exposed to language models. Rather than forcing premature agent adoption, Notion adopted a "Agent Lab" thesis focused on understanding how people collaborate and building product systems around frontier capabilities. The company deliberately times its roadmap to avoid swimming upstream against model limitations while building features ready for when models mature. Notion's approach extends beyond simple model wrapping to fundamentally rethinking how agents integrate into enterprise workflows. The company has organized dedicated teams around core AI capabilities, product packaging, and strategic eval frameworks that intentionally test at the frontier of model capabilities. Sachs emphasizes an objective-driven culture with low-ego teams comfortable discarding work, while Last highlights the "Simon Vortex" approach to mobilizing hackathons and cross-functional security reviews. The vision extends to "software factories"—systems where agents collaboratively spec, code, test, debug, review, and maintain codebases together, representing what some at Notion view as foundational AGI infrastructure.