CHAPTERS
Notion’s goal: become an agent orchestration platform
Eric (a Notion PM working on agents) explains the vision of Notion as the place where humans and multiple agents coordinate work. The emphasis is on orchestrating the right agents for the right jobs inside existing workflows.
The “prototype sprint” moment: letting Claude run long task lists
Eric describes an internal prototype-building experience where he dropped a large set of tasks into a system and let Claude execute autonomously. The speed and completeness of results became a key “unlock” for how they think about agent-powered building.
Why Claude Managed Agents: long-running, complex work without heavy engineering
Eric explains why Notion was excited about Managed Agents: they make sophisticated agent behaviors easier to integrate. He contrasts plug-and-play managed infrastructure with the effort required to build comparable orchestration and persistence in-house.
Integration approach: cloud-run sessions + API hookup
Notion integrates Claude Managed Agents by running agent sessions in the cloud and connecting through the API. This enables Notion users to kick off many concurrent jobs from within Notion without managing the execution environment themselves.
Use case setup: client onboarding for non-technical teams
The demo centers on a common customer workflow: onboarding a new client. Eric introduces a fictional client (Harbor and Pine) and frames the problem as turning scattered onboarding action items into structured tasks inside Notion.
Custom onboarding agent with full Notion context
Eric shows a Notion custom agent that has access to relevant client databases and task boards, giving it the context needed to perform onboarding work. Claude operates behind the scenes, using that in-Notion context to generate outputs.
Orchestration in action: turning action items into tasks via multiple threads
Eric triggers the “client onboarding manager” agent to convert action items into tasks. Notion spins up multiple agent threads that feed into Claude, enabling parallel processing and rapid task creation in the task board.
In-Notion chat that mirrors the underlying Claude session
The workflow allows users to converse with the Claude session directly within Notion. The chat experience reflects the same session that was initiated for the managed agent run, enabling iteration without leaving the workspace.
Developer visibility: viewing the run in the Claude platform and using traces
Eric opens the Claude platform view to show the managed agent session running and describes why this is valuable for development. Traces help the team understand behavior and improve agent quality over time.
What makes long-running agents work: persistence, memory, and output quality over time
Eric highlights that reliable long-horizon execution is essential—tasks may run for 20 minutes or an hour. Capabilities like continuing execution, managing memory, and maintaining output quality are described as critical layers beyond the base model.
Results and iteration: generated onboarding artifacts (e.g., homepage example)
The agent completes many tasks and produces example deliverables, like a homepage concept, using Notion context. Eric notes that users can immediately iterate by continuing the conversation with Claude inside Notion.
Takeaway: Managed Agents as a scalable “playground” for parallel work
Eric closes by emphasizing how empowering it is to launch many jobs simultaneously and treat Managed Agents as a creative playground for building workflows. The overall message is that managed orchestration unlocks speed, scale, and ease of experimentation.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome