What Should I Even Build?

18 months ago Jonny and I opened our first AI workshop by asking everyone to introduce themselves as an AI-generated dessert.

I dug that FigJam board out this week. Lots of it now seems dated. (Although I’m now feeling peckish.) But there’s one exercise that was particularly useful, and it’s what we’re sending to everyone booked onto next week’s Build Your Own Design Tools workshop.

It answers ‘what should I even build?’ by starting with your week rather than with the technology.

Here’s how it worked.

1. Narrate your week

Open your calendar / task list

  • Where are you spending time?
  • What are your recurring meetings?
  • Which blocks of the week energise you?
  • Which do you find draining?

Write everything down, or just voice transcribe for 10 minutes. This is about building rich context of what’s meaningful to you. Yes, you could connect your calendar and do this automatically, but it’d be far less personally meaningful.

2. Sort into three buckets

Ethan Mollick’s a clever guy. He writes smart things about practical everyday uses for AI, and we use his framework to sort everything from step one:

  • Just for me. I don’t want (or trust) AI anywhere near this!
  • AI assisted. AI could probably help me be more efficient or creative here.
  • Fully autonomous. Tedious as hell, please don’t make me do this.

Importantly, this is highly subjective. Pushing pixels in Figma could fall in any of the three buckets depending on the person.

We’ve found this to be an invaluable step for keeping the AI anxiety at bay, and for letting people see the actually quite boring and repetitive parts of their week that they’d like help with.

3. Come up with 100 ideas

This is intentionally over the top. People bring their own ideas of course, but 100 is a LOT.

The volume constraint forces working with AI, often leading to a creative back and forth to reach this arbitrary quantity. Before deferring to human curation to distill down to ten actually decent ideas. (While chuckling at how bad some of the pure AI ideas are.)

The FigJam version took an hour and a wall of stickies. We’ve rebuilt it as a Claude Skill that runs the same three steps in about 15 minutes, and ends with 10 design tool ideas you could build in half a day. It calibrates to your skillset and how ambitious you’re feeling, so nobody gets handed something unbuildable.

Half will be useful. Half will be daft. Take the daft ones seriously though.