Instinct, on X

Today
Andrew Lee
@startupandrew
#1Mon 1:30pm ET

There's been a lot of buzz about — and criticism of — Instinct over the last few days. Building an agent that YOLOs it can produce impressive results. Building an agent that you can actually trust with real data & responsibilities is much harder. At @TaskletAI much of what we do is giving businesses enough control that they can trust agents for real, consequential work. Here's a few things we to make this real

What people said back

Quote-posters framed it as cool demos versus scary permissions and asked whether a CFO would ever hand an agent live financials. One Instinct user said the argument already converted them, while others asked practical questions about locking down patient data.

TechCrunch
@TechCrunch
#2Mon 2:05pm ET

Early testers are raving about what Instinct can do, but some say the AI assistant’s sweeping access, broad terms and ability to act on users’ behalf come with uncomfortable trade-offs t.co/0HRs7HC3EM

What people said back

Replies split between alarm (perpetual inbox rights, regulators, find out in three months) and patience (it is still closed beta, the policy is a placeholder). One quote-poster reported 2,486 messages and 110 completed tasks, unbothered.

Krishna Hegde
@PositiveGamma
#3Mon 2:06am ET

Instinct is so good, I have now pinned it at the top of my Whatsapp. Queries that would have been made on Google 12 months ago and Claude/ChatGPT last month are now making their way to Instinct. Also, the way it helps me stay on top of my different Whatsapp discussions/plans

What people said back

Almost the entire thread was people asking for an invite, plus one reply arguing the real unlock is always-on VMs and self-triggering agents wrapped in a harness nobody has to understand.

Rak Garg
@rak_garg
#4Mon 4:26pm ET

Instinct in the past 15 days has: 1. found and booked me with a new dentist 2. found and booked me with a new in-network acupuncturist 3. found and scheduled an appt with a cobbler for a bunch of things I needed to get fixed 4. deleted 6k marketing and fluff emails 5. sends a daily digest of newsletters w/ just the things I need to pay attention to 6. schedules date nights w/ my wife based on both of our calendar availabilities 7. makes vet appointments for our dog's allergy shots 8. makes padel bookings for me when they open 9. is talking to various dealers / private party sellers for a car I'm looking at buying at the very least this thing is my liaison to the entire medical/healthcare system and I am better off for it

Ayush
@ay_ushr
#5Mon 6:15am ET

I am very okay with giving agents my card as long as it has good confirmation mechanisms (eg with Instinct). I find the experience pretty seamless for me already I'm sure some of these agent protocols like x402 / MPP are better designed in theory, but tbh I'm not seeing a need for them I can't think of any common use case I have that wouldn't work with existing card rails

the funny thing about 'agentic payments' is that they're here, probably doing significantly more volume than x402 or MPP - it's just that it's coming through computer use or CLIs you're signed into already
τop τick crypτo 📁 🤖🧠
@toptickcrypto
#6Mon 9:30am ET

There have been efforts around private inference on TEEs that were highly experimental but I think open-weight models are getting to the point someone has the opportunity to create a verifiably private t.co/n8YoW0Vv8j that could find pmf.

What people said back

Broad agreement that the stack is finally ready, with one reply noting that distribution, not technology, is the actual moat for anything privacy-flavored.