Simon Last was sitting a few meters away, one of Notion’s three founders. The engineer is shy and gangly. He has given up his management duties to concentrate on what he loves most: being an architect. “super IC”—an individual contributor. He stood and shook my hand. I apologized for being awkward. Simon went back to his computer, where an AI was coding for him. Simon would say later that using AI apps to code was similar to managing a group of interns.
Then, in 2022 the The AI assistant of Notion has been helping users to draft notes. This is now being re-branded as an assistant. “agent,” This is a form of AI which will run autonomously and in the background while you are tackling other tasks. Human engineers must write a lot of code to pull off this feat.
Cursor opens and they choose from a variety of AI models. Claude or the Claude Code App was preferred by most of the engineers with whom I talked during my visit. Engineers select their AI and then ask them to create code that will build something new or improve a certain feature. The human programmer then debugs and tests the output as needed—though the AIs help with this too—before moving the code to production.
The core of generative AI, however, is extremely expensive. The savings are in time. For example, Ivan Zhao, cofounder of Notion and CEO, could go to his office’s jazz club and relax for some time if AI completed his task earlier than anticipated. Ivan likes jazz music. He fills his time in reality by working harder. It will not be possible to maintain the fantasy of a four-day week.
The ultimate code sprint, my workweek was only two days at Notion. For full access, I had to agree to only identify engineers using their first names. I was given the task of fixing the appearance of a chart known as a “mermaid” diagram in the Notion application. Quinn Modi told me the diagrams in Notion are SVGs and can not be magnified or scaled, even though they have been called vector graphics. The text in mermaid charts on Notion can be unreadable.
Quinn pulled his laptop in my direction. He was running Claude on the Cursor application. To have some fun, he browsed through Notion’s source code. “So, the Notion code base? Has a lot of files. You probably, even as an engineer, wouldn’t even know where to go,” He said politely, referring me to as an engineer. “But we’re going to ignore all that. We’re just going to ask the AI on the sidebar to do that.”
Quinn’s strategy to code his vibes was asking the AI: Why does this look the way that it does? It forces the AI into doing some initial research, which helps us, as human engineers, to write the next prompt. After “thinking,” Cursor told us via a stream of lines that Notion’s Mermaid Diagrams were static images, which, amongst other things, lacked click handlers, and weren’t integrated into a full screen infrastructure. Sure.
