Post by Haris Mahmood
Founding AI Engineer @ Unify (YC W23)
Excited to see this finally out π The thing I find most interesting about what we've built: it doesn't need you to get the instruction right the first time. You teach it the way you'd teach a person - share your screen and walk it through something live, send a voice note, hop on a call, correct it mid-task. Whatever's most natural. And it retains all of that as structured knowledge - contact profiles, workflow preferences, domain facts - not just conversation history it scrolls back through. π What makes that work is that your corrections and guidance don't just sit at the surface - they thread through every layer of work happening underneath. Redirect something mid-flight and the change propagates all the way down, without starting over. That's a harder problem to solve than it sounds, but it's what makes the experience feel particularly different. Would love to hear people's honest thoughts (both good and bad).
Video Content