Why not try to define a strict subset of the current specs, that would target ease of implementation & graceful degradation? I'd rather have many different clients compatible with a "web-lite" spec that is enough to navigate on 95% of websites, which would have an incentive to officially support that subset if it becomes popular enough.
Thanks for the write up. Always interesting to see how very senior developers interact with AI these days.
@antirez: Introducing a regex feature that late into the project for a seemingly unrelated feature feels a bit weird? Can you explain more your rationale on that? thanks!
I'm using this as my main way to run Claude & others, it's just a shell wrapper but it provides a lot of niceties and a pretty good developer experience in a single command line. Curious to have feedbacks on it!
Love Oxide and what they're building, but I'm not sure raising even more VC money is the way to build a generational company. Quite the opposite? With money you don't need, you're trading faster growth for more dependencies on third parties that will seek a ROI eventually?
This is the way to go! On my side I've build a very small `claude-vm` wrapper to run each instance in a VM with Lima: https://github.com/sylvinus/agent-vm
The Ulanzi TC001 is a great, cheap piece of hardware. I found a second-hand one for $20 and flashed it via USB with https://github.com/lubeda/EspHoMaTriXv2, a more practical firmware if you already have a bunch of ESPHome devices at home.