CHAPTERS
- 0:00 – 0:30
Prototype breakthrough: letting Claude run multiple tasks end-to-end
Eric, a Notion PM, describes a key moment while prototyping agent orchestration: he dropped a large set of tasks into Claude, walked away, and returned to completed prototypes. This experience reframed what was possible with long-running, parallel agent work inside Notion.
- •Eric’s role: agents and agent orchestration at Notion
- •From dozens of tasks to a working prototype quickly
- •The “walk away” moment: Claude continues autonomously
- •Seeing parallel execution as a major productivity unlock
- 0:30 – 0:45
Notion’s vision: becoming the agent orchestration platform
Eric explains why Notion is well-suited for human–agent collaboration and workflow management. The goal is to bring the right agents together for specific jobs and coordinate them across real work artifacts (databases, tasks, docs).
- •Notion as a hub for human + agent collaboration
- •Orchestrating multiple agents for different roles/jobs
- •Managing workflows as a first-class product capability
- •Context-rich workspaces enable better agent performance
- 0:45 – 1:00
Why Claude Managed Agents: complex, long-running work without heavy engineering
The discussion shifts to why Managed Agents are compelling: Claude can handle complex tasks that run over extended periods, and Managed Agents reduce the engineering lift to make that reliable. Instead of building a sophisticated orchestration system from scratch, Notion can integrate a plug-and-play solution.
- •Claude excels at complex, long-running tasks
- •Building equivalent infrastructure yourself is a “mega brain” effort
- •Managed Agents simplify reliability and orchestration concerns
- •Product value: faster iteration and easier integration
- 1:00 – 1:16
Cloud sessions + API integration: how Notion wires Managed Agents into the product
Eric outlines the integration model: Claude sessions run in the cloud and Notion connects via API. For customers, this enables kicking off many jobs directly from within Notion without needing to manage underlying execution infrastructure.
- •Claude runs sessions in the cloud
- •Notion pulls in the API to connect functionality
- •Customers can launch many jobs from Notion
- •Infrastructure abstraction enables product focus
- 1:16 – 1:31
Use case setup: client onboarding for non-technical teams
Eric introduces a concrete example—onboarding a fictitious client (“Harbor and Pine,” a lifestyle brand). This is framed as a common workflow for non-technical users that benefits from an agent that can interpret action items and turn them into structured work.
- •Client onboarding as a representative workflow
- •Designed for non-technical users
- •Example client: Harbor and Pine (lifestyle brand)
- •Action items become the raw input for automation
- 1:31 – 2:02
Custom onboarding agent inside Notion: grounding in databases and task boards
Notion uses a custom agent that has access to the client databases and task boards, giving it full workspace context. Claude powers the agent behind the scenes by pulling relevant Notion context to generate onboarding artifacts and next steps.
- •Custom agent is granted access to relevant Notion data
- •Full context: client databases + task boards
- •Claude operates behind the scenes to synthesize outputs
- •Agent supports end-to-end onboarding workflows
- 2:02 – 2:17
Turning action items into tasks: parallel agent threads feeding Claude
Eric demonstrates the workflow of converting action items into a task board. As the items are dumped in, Notion spins up multiple agent threads that feed into Claude, which then launches a Managed Agent session to execute the work.
- •Action items are ingested and converted into tasks
- •Tasks populate directly into a Notion task board
- •Multiple agent threads run in parallel
- •Threads feed into a Claude Managed Agent session
- 2:17 – 2:33
In-Notion chat mirrors the Claude session: interacting during execution
Notion provides a conversational interface where users can talk to the Claude session without leaving their workspace. This chat mirrors the underlying Managed Agent session that was kicked off, enabling iterative guidance and refinement mid-flight.
- •Chat with Claude session directly inside Notion
- •Conversation mirrors the running Managed Agent session
- •Supports iterative prompting during task execution
- •Keeps work anchored to the workspace context
- 2:33 – 2:48
Developer visibility in the Claude platform: monitoring, traces, and improvement loops
Eric shows the Claude platform view of the running Managed Agent session, which is useful for development and debugging. Traces from execution can be used to understand behavior and continuously improve the agent’s performance.
- •Claude platform offers a different operational view
- •Useful for debugging what’s happening during runs
- •Execution traces provide insight into agent behavior
- •Traces feed back into improving the agent over time
- 2:48 – 3:03
Why long-running harnesses matter: persistence, memory, and quality over time
The conversation emphasizes that long-running tasks may need 20 minutes to an hour, requiring durable execution. Managing continuity, memory, and maintaining high-quality outputs over extended runs is presented as a critical layer on top of the underlying model.
- •Some tasks require sustained runs (20 minutes to an hour)
- •Need continuity and execution management beyond one-off calls
- •Memory management is essential for coherent long tasks
- •Quality over time is a key differentiator for real workflows
- 3:03
Results and iteration: generated onboarding assets and refinements in place
Eric notes that tasks are already being completed and shows an example output such as a homepage concept generated from Notion context. Users can then iterate by continuing the conversation with Claude directly within Notion.
- •Tasks complete quickly once the workflow starts
- •Example output: a generated homepage concept
- •Outputs are grounded in Notion context
- •Iteration happens via direct chat in the workspace
Wrap-up: Managed Agents as a parallel-work playground for building in Notion
Eric concludes that Managed Agents feel like a playground—enabling many jobs to run at once and accelerating building. The overall message is that this capability helps both Notion’s internal development and customers’ real workflows.
- •Managed Agents enable many concurrent jobs
- •Fast experimentation and building velocity
- •Benefits both internal teams and customers
- •Agent orchestration becomes more accessible and practical
