My top 5 takeaways from Tanay, @wisprflow's founder, on building AI products that stick: 1. "We no longer have ARR as a North Star" AI companies can hit millions in ARR from new users who churn fast, then get trapped using paid spend to fuel unsustainable growth. Instead, focus on retention and habit formation—70% of Wispr's users retain every year. 2. To help users form habits, study games From Tanay: "Most software is terrible at building habits. I look at games." He models Wispr Flow after Mario where "playing the game is the onboarding." Users level up instead of getting overwhelmed upfront. 3. Avoid decision by committee ""For every single thing, there is one decision maker. Everybody else just has an opinion." Tanay's 25-person team now runs 20 projects in parallel by assigning clear decision makers. 4. Require AI prep work before every meeting "People need to show up having done the basic ChatGPT work beforehand. We spend the meeting on: Of these 20 ideas ChatGPT suggested, what are the three we're doing?" 5. Curiosity → competency → passion. "Curiosity and competency often precede passion. Once you see yourself killing it at something, you just want to do that again." He learned to code at 9—"I was shit in the beginning, but a year later I was less shit."

The biggest AI unlock for me has been to default to voice instead of typing. This chart from @WisprFlow is a fascinating look at how its users learned this over time.

If you use one of the top AI browsers (Open AI Atlas, Perplexity Comet, Dia) - what's your favorite practical use case so far? I haven't found them super useful outside of getting insights for YouTube videos