From Coder to Manager: Navigating the Shift to Agentic Engineering with Notion Co-Founder Simon Last - No Priors: Artificial Intelligence | Technology | Startups Recap

Podcast: No Priors: Artificial Intelligence | Technology | Startups

Published: 2026-03-12

Duration: 29 min

Guests: Simon Last

Summary

Simon Last, Co-Founder of Notion, discusses the transformative impact of AI on productivity tools, the evolution of Notion's engineering and product teams, and the new role of humans as 'agent managers' overseeing AI-driven workflows.

What Happened

Simon Last shares how Notion began its journey into AI after experimenting with GPT-4 at an off-site in Mexico in 2022. Inspired by the capabilities of the model, Notion initially developed AI Writer as a writing assistant to help users rewrite, edit, and generate text quickly. This feature was launched within months, demonstrating the company's rapid experimentation cycle.

Notion's long-term vision involved creating a general assistant capable of performing complex tasks like generating databases, querying information, and autonomously managing workspaces. While this proved challenging, iterative development led to the release of Notion's personal and custom agents by 2023, enabling users to delegate tasks like filing bugs, managing Slack channels, and performing email triage.

Last reveals that Notion has rebuilt its AI harness every six months to stay aligned with rapid advancements in large language models (LLMs). He credits this continual reinvention as crucial for maintaining robust and high-quality implementations, even as the complexity of AI-powered workflows increases.

The episode dives into Notion’s creation of APIs designed for agents rather than humans, employing markdown dialects and SQLite for efficient communication. These tools allow agents to seamlessly interact with documents and databases, ensuring smoother integrations and autonomous task execution.

Last describes how Notion has embraced chaos internally, with more prototypes and ambitious pull requests becoming the norm. Non-technical teams, such as the People team, have adopted custom agents for workflows, showcasing the accessibility and utility of AI within diverse organizational contexts.

The conversation highlights the growing divide between average engineers and those leveraging AI tools effectively. Last emphasizes how tools like coding agents have made it possible for individual engineers to achieve outputs 100x or even 1000x greater than before, raising the stakes for adoption.

Finally, Last outlines Notion’s role as a “Switzerland” for AI models, integrating both proprietary and open-source options while remaining a neutral platform for human-agent collaboration. He envisions agents becoming increasingly autonomous, bootstrapping capabilities like integrations and connections to new data sources.

Key Insights

• Notion rebuilt its AI harness every six months to keep pace with rapidly evolving large language models like GPT-4. This constant reinvention avoids stagnation and ensures AI tools stay effective even as workflows become more complex.
• Notion created APIs specifically for agents, not humans, using markdown dialects and SQLite to streamline communication. By optimizing for autonomous task execution instead of manual input, these APIs make integrations frictionless.
• Internal chaos at Notion has fueled innovation, with prototypes and ambitious pull requests becoming routine. Even non-technical teams like HR have adopted custom AI agents, proving that advanced tools can break out of engineering silos.
• Coding agents at Notion have amplified individual engineers’ productivity by 100x or more, widening the gap between those who embrace AI tools and those who don’t. The stakes for staying relevant in tech are higher than ever.

Key Questions Answered

What is Notion AI's custom agent and how does it work?

Notion AI's custom agent, launched in 2023, allows users to create autonomous agents tailored to specific tasks. Users can grant the agent access to Notion databases or external tools like Slack, enabling it to file tasks, triage feedback, and even respond to messages.

How has Notion rebuilt its AI harness over time?

Simon Last explains that Notion rebuilds its AI harness approximately every six months to keep pace with advancements in LLMs. This iterative process ensures the harness aligns with the latest capabilities and maintains high-quality implementations.

What tools does Notion use to make its APIs agent-friendly?

Notion redesigned its APIs to use markdown dialects and SQLite, enabling AI agents to interact efficiently with documents and databases. This shift ensures a seamless and intuitive experience for agents while reducing token usage and improving performance.