Comment by dvt

We're still 6-12+ months away from a "killer" AI product. OpenClaw showed what's possible-ish, but it breaks half the time, eats tokens like crazy, and can leak all kinds of secrets. Clearly there's potential there, and a lot of people are working on products in the AI space (myself included), but anyone that's seriously tried to wrangle these models will agree with the reality that it's very hard to reliably get them to do what you want them to do.