Dax opens the interview with a paradox: he co-built one of the most-used AI coding tools, uses it daily, and still doesn't feel meaningfully faster.

The core tension

  • OpenCode was built for the team's own use — "the easiest products to build are ones that you can use yourself" — and it is a critical part of their workflows 3:23.
  • Yet the old problems remain: "I'm working as hard as I ever have. I'm struggling as much as I ever have" — objectively things got easier, but the thinking load hasn't dropped 3:46.
  • No competitor in the coding-agent space — where everyone is maximally into AI — is crushing the others through better AI usage. If AI were a decisive advantage, this space would show the gap, and it doesn't 6:28.

Why speed doesn't translate into better products

  • Impact of AI differs by company lifecycle: pre-product-market-fit, at PMF (OpenCode's stage), and decade-old PMF companies. Pre-PMF, AI barely helps because the job is figuring out what to do — "it's better to think a lot instead of swinging a lot" 4:37.
  • At PMF the temptation is 1:1 reactivity: user problem → prompt the agent, competitor feature → prompt the agent. Shipping a thousand features that way "actually adds up to a horrible product" — nothing cohesive, a Frankenstein 5:18.
  • Shipping is a one-way door: "The moment you ship something, you're stuck supporting it forever" — every future feature interacts with it, so restraint still matters 5:41.
  • Being able to ship 10× more doesn't mean having 10× as many good ideas. Dax's current struggle is how to slow everyone down 6:04.
  • The last 6 months OpenCode operated very differently from before; a lot went wrong, and they're now pulling back to see what from the "old world" still makes sense 6:28.