Speed and Quality Aren’t a Trade-Off

Randy Hunt has spent the last few years as Head of Design at Notion. He also chairs the MFA Design programme at School of Visual Arts.

Whew, he’s a busy guy!

Jonny and I sat down with him to talk about building agentic workflows at Notion, leading the team designing experiences around them, and how he thinks about preparing the next generation of design talent.

My takeaways:

  1. Speed and quality aren’t the trade-off we tend to assume they are. For Randy, quality is “the number of times an end user experiences the thing being bad or good”, so with 10,000 people using it this week, 11,000 the next and 12,000 the week after, every week a known fix waits is thousands more encounters with the worse version.

  2. Talk to your printer before the job goes to press. His model for killing the handoff between design and engineering comes from print production. The printer tells you what a scoring decision will cost, mentions the materials already sitting in the warehouse, and then you physically go and stand at the press when printing starts to watch the first sheets roll off. Passing the baton with no interaction either side is “like playing a game of telephone, but with far more variables.”

  3. A Notion designer was once told a piece of implementation work would take four weeks. They didn’t believe it so they built the whole thing over a long weekend in Claude Code, micro-interactions and all. It landed as one giant PR, and the engineers said reviewing it would take two weeks. When making gets cheap, the bottleneck moves to review.

  4. His team shares their agentic work openly within the company. So everyone can see and learn from how colleagues are prompting Claude. Then, when someone spots a pattern, they’ll ask what commands were used, write it up, and present the doc at the next weekly design meeting.

  5. On the boom in homegrown internal tools, Randy is candid that some of it is “I have a job interview, I need to show people I’m AI savvy, so look, I built this tool.” He expects it to normalise the way open-source packages did, where a handful of good ones do the job well. Nobody writes their own web typesetting tool any more after all.

  6. AI literacy will end up sitting beside visual literacy and media literacy. SVA’s MFA is built as a barbell with durable foundations like ethics and intellectual property at one end, and containers loose enough to swap the contents at the other. Randy has AI literacy at the volatile end for now, and expects it to settle into the foundations fast.

If you’ve already listened to the ep, anything you’d add?

Listen at nearfuture.works/podcast.