Enterprise Singapore X Notion AI Workshop

An evening workshop with Enterprise Singapore taking a mixed audience, mostly Notion beginners with a handful of intermediate users, from Notion AI fundamentals all the way to autonomous agents and AI computer use. I presented alongside Ernest (Notion and Notion AI basics) and Dennis (custom agents and Notion Workers), with live demos throughout and a demo workspace attendees could keep experimenting in for two weeks after the session. The whole event ran with AI meeting notes transcribing live in the background, which became a demo in itself.

What I covered

• AI skills as reusable instruction pages. How a skill is just a Notion page of structured instructions the AI reads before acting, replacing manual prompt engineering with a "save this as a skill" workflow.
• A team skills database. How we organise skills with owner and usage metadata, embed the database in each person's instruction page, and have the AI scan short descriptions on demand to avoid context window overload.
• A real example. My Setup Call Discovery Prep skill, which researches the contact, drafts discovery questions and prepares the pre-call email in one prompt.
• Computer use and HTML generation. Notion AI's sandboxed computer environment, building an HTML presentation live from a mock project proposal, and embedding the result straight into a page with the /html block.
• Shipping real products from Notion. The NasriFlix story (PRD in Notion, HTML prototype, Next.js app, deployed via GitHub and Netlify) and Nas Bobble, a browser game hosted inside a Notion page.

Key takeaways

• Skills beat prompt engineering. Write the workflow once, share it with the team, and everyone gets consistent outputs for proposals, emails and content.
• Custom agents work while you sleep. Trigger them on a schedule or an event, give them the minimum access they need, and audit them through the Insights tab.
• Notion AI is strongest when Notion is your context layer. It outperforms connecting an external LLM via MCP because it is post-trained on Notion search and content relationships.
• The PRD workflow avoids context degradation. Use one long chat for design decisions, distil it into a concise PRD, then start a fresh session to build.
• Notion Workers complete the stack. Cheap scheduled syncs pull external data into databases, pairing with charts and dashboards for a lightweight BI setup inside Notion.

Outcome

• Attendees left with two weeks of hands-on access to the demo workspace to build their own agents and skills.
• Dennis built a working daily email triage agent live, and the audience followed along building their own.
• A full HTML presentation was generated live from a proposal document using computer use.
• Strong Q&A covering custom writing styles, agent safety and permissions, second brain systems and Notion AI versus Claude, with questions answered on the spot.

Project deck