Since starting my role at Lightspeed, I’ve become AI-pilled, and spend a not-insignificant part of my day keeping up to date with what’s going on with the various labs. Inspired by a post from my friend Shrivu (who is far more cutting-edge than I am), I thought I’d quickly share which tools I use:
My default go-to model for most things is Claude Sonnet 4. It’s hard to explain why exactly, but the vibes of using Claude are just solid; whenever I have a quick question that pops up in my head I reflexively open up the Claude app. It performs especially well at work, though I’m a bit cautious of a bit too much agreeableness and verbosity - Claude can be a yapper. I have both a Pro ($20/mo.) subscription for personal use and an Enterprise subscription through work.
My image generation model is ChatGPT (GPT-image-1). It was such a step up when it was first released and I’ve noticed its work in the wild walking around San Francisco - it has a signature cartoon-y aesthetic that I really enjoy. I’ve also been impressed with its ability to read text in photos in other languages - there’s a Thai restaurant I go to here with only Thai writing on the daily specials blackboard, for whatever reason I just trust ChatGPT more to do the job.
Lastly, my voice conversation model is ChatGPT (GPT-4o currently, though I imagine it’ll migrate over to GPT-5 soon). It does a much, much better job of (1) turn detection, (2) voice quality, and (3) providing visual feedback in the app versus Claude. Talking with Claude I frequently run into Claude cutting me off or Claude no longer listening to me; ChatGPT nails it, responding quickly and showing what it’s saying on the screen. I also just find the audio it generates more natural sounding (whereas Claude sounds incredibly robotic).
I don’t write code very often, but were I to do so I’d use Claude Sonnet 4. Anecdotally if I ask someone which model they use for most things, there’s an extremely high degree of correlation between someone working as a software engineer and someone using Claude as their default model.