Notion's New AI Agents for Work Blew Me Away (Full Tutorial) | Akshay & Ryan
Everyone calls their AI product an “agent” but Notion’s agents actually live up to the hype. This is the story of how 2 years of failure led to building the best AI agents for work.
Dear subscribers,
Today, I want to share a new episode with Akshay Kothari and Ryan Nystrom.
Everyone calls their AI product an “agent” but Notion’s actually live up to the hype. It took Akshay (Notion’s co-founder) and Ryan (AI lead) 2 years of failure to build agents that can create databases, respond to Slack, and more. We had a great chat about their most surprising lessons from building the best AI agents for work.
Watch now on YouTube, Apple, and Spotify.
We talked about:
(00:00) What it really took to build agents that actually work
(01:38) Live demo: Watch a database build itself in real-time
(04:56) How Akshay writes docs by just talking to AI 90% of the time
(20:15) The real difference between AI slop and useful AI agents
(34:43) Personalizing agents with custom memory and instructions
(39:59) Live demo: Building a custom Notion agent for Slack
(49:03) What Notion looks for when hiring AI-native builders
This episode is brought to you by…Bland
Bland creates AI customer support agents that sound human, speak any language, and work 24/7. Features include custom prompts keep conversations on-brand and accurate, integrations with your CRM, and self-hosted infra with 99.99% uptime.
Companies like Gallup, Samsara, and Snap trustBland to handle millions of interactions a day. To try Bland, call the number in the image above or…
Top 10 takeaways I learned from this episode
Structure your data the way AI models naturally consume it. Notion’s agent breakthrough came when the team stopped trying to get the AI model to understand their internal page structures and instead presented everything as markdown (a format that AI already understands) for the AI model to consume.
Focus agents on eliminating busywork, not building toy demos. The turning point for Notion was “when we decided to stop building toy apps that look like AI slop and instead focus on actually helping you get your work done.” Don’t chase shiny AI trends—solve the annoying manual tasks users actually hate doing.
