hey X ๐
I'm Aditya, doing growth at an AI startup @beanchors_
also building something in stealth for marketers tired of guessing which tool to buy ๐คซ
here for growth, AI & startup convos
what are you building? let's connect ๐ค
@currocreates The one-line constraint is a great clarity filter ๐ For your free tools for patients, who ended up using them that surprised you most?
@_avichawla This is why temp 0 still isn't really deterministic ๐คฏ Your eval run quietly depends on everyone else's traffic. Do you think providers will start selling a batch-invariant "deterministic mode" as a premium tier?
@Afield_Edward@wunderwuzzi23 Agent security is going to matter a lot more as agents get real permissions.
What was your biggest takeaway from talking with him, prompt injection or something else? ๐ค
@DhravyaShah Customers holding their own data while you serve the models feels like the unlock for regulated teams ๐
Do you think most would want it fully self-hosted, or is a VPC deploy enough?
Question for people building with AI:
When you ship an AI feature, how do you tell if it's actually helping users? Usage, retention, thumbs up/down, or fewer support tickets?
Curious which signal you trust most.
4/ Re-test it as a stranger.
Watch one person who has never seen the product try it without help. Where they pause is where your users quit.
What's the one onboarding step you'd cut first in your product?
3/ Cut, delay or pre-fill.
Does this step have to happen before the first win? If not, move it after.
Team invites, profile setup and settings can usually wait. Sample data beats an empty dashboard.