Key Takeaways

  • Notion built an MCP connector and immediately faced a massive influx of automated queries from external agents.
  • Tibo Sottiaux predicts the majority of future actions across the internet will be taken by autonomous software agents rather than human users.
  • Developers are making a systematic error by hardcoding complex workflow graphs around current latency and cost limits instead of preparing for 10x model improvements within a year.
  • Product teams must choose between building high-throughput machine interfaces or investing exclusively in rich, human-centered experiences.

The Trap of Present-Day Limitations

Most software teams build for the models they have on their screens right now. They look at current latency numbers, calculate token costs, and assume those constraints will stay fixed. Then they spend months building fragile workflow graphs and hardcoded guardrails to compensate for model weaknesses.

Tibo Sottiaux, Head of ChatGPT and Codex at OpenAI, sees this as a fatal blind spot. Speaking at DevDay around the launch of Dots (OpenAI's autonomous personal agent platform), Sottiaux pointed out that developer assumptions lag behind the hardware and model curves.

“Often when I look at what people are building out there, it's just like, 'You're not quite getting it.' Like, if you were just pushing yourself and just really imagining all of this being roughly 10 times better than it is today in a year, you would build in a different way.”

When models get 10 times faster, 10 times cheaper, and combine text, vision, and audio natively, manual workflow scaffolding becomes technical debt overnight. The complex pipelines teams build today to manage context windows or retry failed API calls will be obsoleted by raw model capability.

The Notion Signal and the Machine Web

The shift from human browsing to agent traffic is not a distant theory. It is already hitting production infrastructure.

Sottiaux highlighted Notion's integration with the Model Context Protocol (MCP). When Notion exposed its data through MCP, external agents immediately began calling those endpoints to complete tasks on behalf of users. The result was a dramatic spike in programmatic traffic that looked nothing like normal human software usage.

“The majority of actions on the internet will be taken by agents. If you want your product to be successful for agents, you have to build for a certain level of scale.”

This shift creates a clear tension for product builders. Human interfaces require visual clarity, deliberate click paths, and forgiving error states. Agent interfaces require high-throughput APIs, predictable JSON structures, and strict authentication designed for thousands of automated requests per minute. As Sottiaux observed, “There's this tension between, if you're building products, do you build an interface or not? And you can kind of hold that back for a while, but it is inevitable.”

What to Do With This

Audit your core application this week and list the top three actions users perform manually. Expose those three actions via an MCP server or a lightweight, authenticated API endpoint, then run an automated agent loop against them to verify whether your rate limits, auth tokens, and database queries survive a 50x spike in programmatic usage.