Q&A: Notion Product Manager Eric Liu
Notion product manager Eric Liu explains how his team built agent orchestration with Claude managed agents.
Notion builds agent orchestration where teams and Claude collaborate on real work
Industry:
Software
Company size:
Large
Product:
Claude Managed Agents
Location:
North America
30+ concurrent agent tasks from a single task board
with Claude handling coding, artifact creation, and client deliverables
90% cost reduction with prompt caching
while cutting latency by up to 85%
Notion is a collaborative AI workspace where teams and agents work together, used by millions of people from individuals to Fortune 100 companies. Claude powers AI features across the product, helping users write, search, and automate data entry. Now Notion is also building agent orchestration with Claude managed agents, letting teams delegate real work, from coding to client deliverables, directly from their task boards.
With Claude, Notion:
- Helps large enterprises like Osaka Gas reduce time spent searching for information by 35%
- Reduced costs by 90% and latency by up to 85% with prompt caching
- Saves customers like Remote an estimated 10 minutes per search across 300 daily queries
- Helps companies like dbt Labs save over $35k annually by eliminating the need for additional AI tools
- Teams can kick off 30+ concurrent agent tasks from a single task board, with results routed to the right people for review
- Supports artifact creation across formats, including code, PowerPoints, PDFs, and websites, for both technical and non-technical teams
The challenge
Q&A: Notion Product Manager Eric Liu
Notion product manager Eric Liu explains how his team built agent orchestration with Claude managed agents.
Q&A: Notion Product Manager Eric Liu
Notion product manager Eric Liu explains how his team built agent orchestration with Claude managed agents.
From finding information to doing the work
Notion users generate enormous amounts of knowledge: meeting notes, project docs, process guides, product specs. The first challenge was making it all findable. Customer support teams need troubleshooting steps. New sales reps need Salesforce processes. Product designers need brand guidelines. Before Claude, answering these questions meant searching across pages manually or waiting for a colleague who knew where to look.
"We want to automate busy work out of knowledge work and free up humans so they can do more high-value activities," said Simon Last, co-founder at Notion leading its AI efforts.
But as AI agents became capable of producing real work, a second challenge emerged: most agent interactions are one-to-one, a single person working with a single agent on a single machine. There's no visibility into what agents produce, no approval workflows, no way for colleagues to step in and iterate together. "People are building super powerful agents for every vertical slice of work, from customer support to coding," said Eric Liu, Product Manager at Notion. "Why don't we bring it all into Notion?"
The solution
Claude Managed Agents: Get to production 10x faster
We're launching Claude Managed Agents, a suite of composable APIs for building and deploying cloud-hosted agents at scale.
Claude Managed Agents: Get to production 10x faster
We're launching Claude Managed Agents, a suite of composable APIs for building and deploying cloud-hosted agents at scale.
Claude Managed Agents: Get to production 10x faster
We're launching Claude Managed Agents, a suite of composable APIs for building and deploying cloud-hosted agents at scale.
Selection Claude for agent workflows
When evaluating models for its AI features, Notion looked for quality that matched the product's standards. Last's team tested multiple providers and found Claude Opus consistently ahead on the criteria that mattered most: reasoning, instruction-following, and tone.
"Claude powers many use cases where we really care about the response's feel and vibe," Last said. "Claude's model quality really impressed us, especially as we wanted to expand to one-click skills where certain tasks like writing and editing were better served by Anthropic's models. Ultimately, we found Claude to be the best across the board."
As Notion expanded into agent workflows, Claude's strengths proved equally relevant. "We've found that Opus 4.6 excels at interpreting what users actually want, producing shareable content on the first try," said Sarah Sachs, AI Lead Engineer at Notion. "Combined with its speed, token efficiency, and surprisingly low cost, it's the first time we're making Opus available in Notion Agent."
“We integrated Claude Managed Agents, which can handle long-running sessions, manage memory, and deliver high-quality outputs over time, to make that possible.”
Product Manager, Notion
The outcome
Notion builds agent orchestration on Claude Managed Agents
Notion's customers made the case for natively building agents within Notion’s workspace. After a demo showing agents working on a kanban board went viral, customers began building their own integrations, rigging together webhooks and custom servers to get Claude running inside Notion. "We saw that customers were willing to jump through hoops to have a native experience of agents within Notion," Liu said, "and Claude was the one people wanted most."
Claude Managed Agents gave Notion a way to offer that experience natively. "We want Notion to be the best place for teams to work with agents and get things done," Liu said. "We integrated Claude Managed Agents, which can handle long-running sessions, manage memory, and deliver high-quality outputs over time, to make that possible. Our users can now delegate open-ended, complex tasks, everything from coding to generating slides and spreadsheets, without ever leaving Notion."
The result is a workflow where teams create a task, move it to "ready to start," and Notion invokes a Claude session. Claude picks up context from connected pages, design systems, API docs, and product requirements documents, then gets to work. For engineering teams, this means prototyping: a product manager records a meeting note, Notion transcribes it into action items, and Claude writes the code. For non-technical teams, Claude generates client deliverables: presentations, brand strategy decks, sample websites.
"Our agent is good at using Notion, but it's not the best at coding, and Claude is very good at that," Liu explained. "And there's a whole world of files beyond Notion pages. Claude is really good at generating that long tail of artifacts."
The workflow isn't one task at a time. Liu described kicking off 30 prototype tasks at once, leaving, and returning to find them all completed. "We turned what used to be solo work into a collaborative moment," he said. The completed work also feeds back into Notion's shared skills database: Claude identifies lessons from finished tasks and updates the skills that future sessions draw from, so agent performance improves over time without manual maintenance.
Claude powers AI features for millions of users
Alongside agent workflows, Claude continues to power the AI features Notion's users rely on daily. AI Writer helps users create and refine content. Autofill populates database fields automatically. And Enterprise Search lets users ask questions about anything in their workspace and get answers drawn from their entire knowledge base and connected apps.
"We've integrated our retrieval system throughout the platform, including in our writing assistance," Last explained. "This allows users to instantly access relevant information from their entire workspace, whether drafting a document or asking a specific question."
The results at enterprise scale have been significant. Osaka Gas estimates a 35% reduction in time spent searching for information. Remote saves an estimated 10 minutes per search across 300 daily queries. For companies like dbt Labs, the comprehensive functionality has eliminated the need for separate AI tools, saving over $35k annually. On the infrastructure side, prompt caching reduced Notion's costs by 90% and latency by up to 85%. "Prompt caching makes Notion AI faster and cheaper, all while maintaining quality," Last said. "This enables us to create a more responsive user experience for our customers."
The AI assistant has proven especially valuable for onboarding. "When someone new joins the company, they use it 10-20 times daily," Last shared. "We get consistent feedback that it's amazingly useful for the first few weeks."
Looking ahead
Notion is building toward a workspace where the same tools that organize human collaboration, task boards, suggested edits, version history, also serve as the interface for working with agents. "Approval flows, suggested edits, all those things that humans already understand," Liu said. "We are the translation layer to agents." As Claude's capabilities advance, Notion plans to expand with broader data sources, additional connectors, and deeper agent workflows that extend across both technical and non-technical work. Liu added: "We want to be the place where people and agents meet to do work."
“We saw that customers were willing to jump through hoops to have a native experience of agents within Notion, and Claude was the one people wanted most.”
Product Manager, Notion