Instinct is working on a desktop app that can read and send iMessages, and use local web browsers Instinct is also working on WHOOP integration for health data reading

I first met @matanSF following a kind intro from @MattEvantic.   We went for a walk in Hyde Park. I was in my trusty short shorts and it was kinda awkward as a walk.  It was awkward because after 5 mins, he was clearly a genius and for the next 55 mins, I just had to pretend like I had more questions to ask before saying with intense eagerness, “can I invest”.  Thank the lord he let me. And through that I got to spend time with his co-founder, @EnoReyes.  Eno is this insane combination of a truly brilliant technologist with an intense awareness of what it takes to build an insanely high-margin, efficient business in AI.  I sat down with Eno when he was in London recently (episode in comments) and have added my handwritten notes below.  Special thanks to @rabois @shaunmmaguire @byersblake @Sabina_Smith_ @laurenmhreeder for some amazing question suggestions. -- 1. The Frontier in AI Right Now The true frontier of AI is building verification systems where no benchmarks exist. Creating systematic frameworks for what “good” looks like allows AI to reliably execute complex, high-friction human tasks that were previously difficult to automate. Is this the true beauty of Instinct @noahrshinn @saranormous 2. The Two Ways That Large Language Models Will Win Frontier model providers face two paths: dominate infrastructure through high-volume inference or move up the stack into high-margin applications. But model-locked applications can conflict with what enterprises actually want: the best possible outcome across multiple models. Love to hear your thoughts on this specifically @AnjneyMidha @mmurph 3. What Is Required to Get the Most Out of Models? Simple gateway routing outside the execution layer delivers only basic cost savings. Maximizing agentic performance requires operating statefully inside the workflow itself, dynamically understanding context, execution history, and what needs to happen next. How do you think about this @alexatallah @shensi @ThibaultJaigu @rauchg @koblovinamerica 4. Why 80% of Neo Labs Will Die and What Separates the Winners From the Losers The winners will anchor themselves to durable enterprise workflows that do not disappear as underlying frontier models improve. Single biggest advice to VCs on investing in neolabs today @LiamFedus 5. Why the Harness Is So Valuable and Who Ultimately Is the Sovereign of Your Intelligence Continuous learning and workflow optimization happen at the harness layer, not inside closed model APIs. True enterprise sovereignty requires owning your harnesses and learning loops so critical intelligence remains proprietary rather than being ceded to third-party labs. 6. Why We Should Not Be Scared to Use Chinese Open-Source Models Labeling open-source weights as dangerous “Chinese models” can obscure the distinction between model provenance and actual security risk. Open models can reflect creator biases, but their risks should be evaluated technically rather than by origin alone. As open weights improve, they could power a growing share of standard enterprise workflows.

You can prompt inject the instinct founder. I just found out that instinct has a feedback tool. I'm 100% sure he will put the feedback just into his coding agent. So you can prompt inject his coding agents.

Very insightful article on agent memory architectures and exploring how Instinct (might) be architected The key point - forgetting nonessential things is just as important as remembering essential things t.co/1oaAnmXhzg

Instinct thought doing this was a dumb idea and I literally had to force it to make it for me, but here’s my 3D bot office

We need more products that are free to the consumer for AI to take off with the mainstream This was the traditional consumer path pre-AI, but because of high COGs everyone launched a subscription I asked Instinct how much I spent through it in the last day - it’s meaningful! t.co/E2SmAatceH

A hot new AI agent called Instinct that reads and answers messages and books flights was valued at $500 million earlier in August, sources say. Now its valuation has skyrocketed to $2.5 billion in another new round just a few weeks later. Read more: t.co/bvRxniWMOC

best way to explain instinct to a friend: it's a sneaker bot but for the rest of your life t.co/KHS1FzVcRq

Instinct is absolutely nuts. First time I’ve felt like the ai personal assistant thing clicked for me. It booked a hotel for me and my family at a discount that I couldn’t find advertised and corresponded with the front desk to get a crib in the room and early check-in. Gronk bot is cool too but Instinct did this 30 seconds after I joined and I didn’t set up any connections or plugins. It just works immediately. So magic.

Starting using Instinct last night. Magical experience booking a flight for my son and making dinner reservations. As a consumer offering, I think it’s the best AI product since ChatGPT launched in 2022. Congrats @noahrshinn!

ok I'm impressed with instinct so far. I think something like iMessage + little web apps will go a long way. The vault experience was really nice.

been testing instinct. it's cool but it's basically a way more powerful poke. same ux, but not as fun to talk to. grok bot on the other hand feels new and fresh. the whole multiple bots thing + channels like slack. it's like having a team that can replace codex/claude for a bunch of things. still worth trying instinct if you haven't. here's a few invites t.co/RKw4mVKU2z

I got access to Instinct, I’ve got 5 invites Who wants a invite?

Stripe just added Link to Instinct, the red-hot AI agent in iMessage & WhatsApp whose buyers spend over $1,300 a month through it. Instinct is the invite-only agent you text on iMessage or WhatsApp that has every VC on Twitter excited. It books flights. Buys tickets. Cancels subscriptions for you. The company closed a $250M Series B this week at $2.5B, up from $100M a few weeks ago, and it's pre-revenue. What makes it different from ChatGPT is that it's really smart and in many cases it "just works" like magic. I saw one user whose agent got them IMAX tickets for The Odyssey by waiting until 3:09am for a cancellation and grabbing it. I've used it to book restaurant reservations and haircuts for the kids. It found a free slot in my calendar, booked it, and dropped in the diary invite. Lovely! I've also watched it get blocked by merchants and fail to authenticate. Things like Instinct work like magic until they don't. Under the hood, you used to store your card details with Instinct, and it would try and do 2FA for you using your email. Which left it open to phishing. Enter @stripe: Now when Instinct buys something for you, Link mints a one-time virtual card after you approve. The agent never sees your real card. The card was the scariest thing you could hand it. Stripe just took that out of the agent's hands. This is smart by Stripe. Link is becoming the wallet agents default to. It already works with OpenClaw and Claude, and on the merchant side Stripe handles checkout inside ChatGPT, Gemini and Copilot. Stripe now sits on both ends of an agent transaction. We're still early. But consumer agents are coming. And most companies are NOT ready. Instinct and Grok Bot point to the same place: every consumer gets an agent, and the agent is the channel. That changes marketing, payments, servicing, the whole relationship. Web to mobile still had a human reading your page. Now the agent reads it, fills your forms, and hits your bot detection. Your CAPTCHA is about to start blocking your best customers.

Instinct is the 2nd company after Poke now to figure out the taste in personal agents been loving it using past few days!!

We need some way to give agents one-off access to our accounts For example I’m asking t.co/bllXmY8EAY to check up on an online I placed for which it needs login access But no way I’m giving it my credentials Just want to give it a one-time access to the site with a limited scope Maybe @1password can build this? Could even be built in a way where sites don’t need to support it. 1Password already has my credentials, so it could do the sensitive part of logging into the site, executing the necessary actions, and then returning the result to the agent. The agent would just tell 1password what to do, 1password checks with me, then does it and returns the result to the agent. Instead of 1password, it could also be a trusted agent running on my own computer. One that executes sensitive tasks on behalf of less-trusted agents

growth experiments that I tried this month with @cmdk_email: 1. Added FAQs to every blog post and landing page, and my impressions in AI overview shot up. Unsure if this would lead to any improvement in domain authority though 2. Cold emailing people with content on email tools, offering them a free lifetime account and a discount code for their audience (with affiliate fee) - Grokbot and Instinct have been great at finding people and their emails

instinct is changing my life. i take back everything i've ever said about ai assistants

Has anyone got an instinct invite they would like to share?

"Memory is a compiler, not a database." @ashwingop's Instinct thesis is the best thing I've read on agent memory — and it cost me a weekend of spec work. I audited KIP, an open cognitive-state protocol I've been building, against his four-layer model. It held on the ledger, on provenance, on belief revision. It failed on one axis: everything was pull-based. Nothing in it represented which change deserves attention. Four things landed in the spec because of that essay: → Watch — durable attention state, evaluated against the commit change stream. Including silence watches: fire if nothing arrives by Thursday. → action_gate — act / ask / defer / silence, recorded as provenance. His sharpest line is that the deepest evidence of a utility function is the silence. Silence you can't audit isn't evidence. → LIST DEPENDENTS — his ghost-inference problem, one layer down. Revoke a root and everything compiled from it becomes discoverable instead of quietly stale. → PURGE PAYLOAD — destroy the observed bytes, keep the evidence record, its digest and its citations. Data minimization without falsifying provenance. One place KIP takes a different path: there is no mutable belief map. Belief is a projection over attributed assertions, computed at read time. "Actually, I'm vegetarian now" needs no patching — his prediction (b) comes free, and the superseded preference stays auditable. And on the moat: if memory really is the moat, that's the argument for making it a protocol. KIP memory leaves as a signed, inspectable capsule. The better the notebook, the less anyone should be able to hold it hostage.

not sure if Instinct is down or I finally gave it an impossible task (US Open tickets)

still don’t know what instinct is and it seems like I hopefully won’t have to find out

If anyone else got confused, Instinct and Poke are the “text it like a person” layer. OpenClaw is the self-hosted OS. Grok Bot is the managed teammate with its own cloud computer. Consumer convenience vs. messaging-native admin w/ full ownership vs. or always-on workers. Table via @grok

Instinct = way smarter Poke = way faster

I do think text agent products like Instinct could be the normie openclaw/agent moment. t.co/pxgWjNqwwF

Instinct is so good

During a delay in Jackson Hole, @JasonYanowitz texted Instinct and it: - Booked five different flights - Refunded the four he didn’t take - Had a car waiting when he landed at 5 AM He didn’t open an app or touch anything else. We're dangerously close to the “personal assistant in your pocket” AI promised.

Lately, Instinct is one product with insane time to magic and delighted me through and through. I'm so far using it only as an email companion but it automagically knows the things I generally delay attending to and surfaces that. Clean product! t.co/0pQ72UgBcA

We’re traveling and staying with friends at the beach this weekend. It’s also my fiancée’s birthday! I wanted to surprise her with something fun and nice, but I have no previous context of the area. So, I asked Instinct to order birthday balloons Nailed it.

Shoutout to @aleks_sharik for inspiring me to use Instinct to help me get my personal gmail to inbox 0 ‍

Using ChatGPT as a real human and not as an employee this past week has been really eye-opening. The amount added via plugins, connectors, and the general UI since I used it a lot in college is crazy. Like, what do you mean you can just change my entire Japan itinerary because I want to go somewhere quieter? Or take photos of some niche torii gate? Oh, the tide is too low , here is the timetable and train routes. Woke up late? Here are the best breakfast spots as options on your calendar. Personal agent companies like @interaction (now @cognition I guess) , instinct, etc. will likely just make a ridiculous amount of money off this market.

Some webdesign inspiration from the world's greatest entrepreneurs t.co/tjfD6Ics8M (Raised $350M at $2.5B valuation) t.co/CtAfTsT18y (Elon Musk) t.co/2Pscx2ftfL (Warren Buffett)

I think the actual value-add of Instinct is that I don't have to interact with mobile sites anymore

I’ve been stress testing t.co/hodEFo5zAJ for the last few days with very normal life stuff: booking restaurants, finding a haircut, sorting little tasks I’d usually procrastinate on honestly really impressed so far what I like most is that it doesn’t feel like chatting with an AI, you ask for something, then there’s this natural little wait while it actually goes and tries to do it, and it comes back only when it has something useful feels much closer to having a real assistant still testing it, but this is one of the first agents where I’m actually finding myself wanting to delegate more stuff to it I still have 10 invites to use, anyone interested?

using instinct solely to snag 10:15 pm reservations at hyped up nyc restaurants

I’ve been using instinct for a while now. It’s amazing. Congrats @noahrshinn