Notion sandwiched between Anthropic and OpenAI

Notion is trying to win at the application layer while renting its intelligence from companies that also want to own the end user relationship. That creates a structurally squeezed position. Anthropic and OpenAI supply the core models inside Notion AI, they also plug into Notion as External Agents, and they sell their own agent products, Claude Cowork and Codex, that chase the same everyday work of drafting, research, and task execution across documents and tools.1 2 3 4 5 6

• The supplier relationship is real and concrete. Notion uses Anthropic to power Notion AI, and its current AI monetization depends on reselling model usage through credits. That means the cost of the core input is set by outside labs, while Notion has separately committed to pushing per workflow costs down, which narrows resale margin over time.1 2
• The partner relationship matters because agents need app level context and actions. Notion is positioning itself as the workspace where outside agents can read pages, update tasks, and act on shared company knowledge. But the same labs can use that access to make their own assistants more useful inside the exact workflows where Notion wants to be the default interface.2 3 4
• The competitive overlap is no longer theoretical. Internal interviews and adjacent research increasingly group Notion, Claude Cowork, and OpenAI Codex together as rival homes for knowledge work. This is the same pattern seen elsewhere in AI, where model labs move upstack from API supplier to full product, like OpenAI pushing from model access into end user software.6 7 8

This heads toward a market where the durable winners either own proprietary workflow context or own distribution at massive scale. For Notion, that makes the next phase about becoming the system where company knowledge and agent actions actually live, so even if the underlying models change, the work still flows through Notion.2 3 4 8