Authenticator support in Instinct Instinct can now securely and seamlessly handle the “enter your 6-digit authenticator code” step for you. It uses the same time-based one-time password (TOTP) standard as Authy, Google Authenticator, Microsoft Authenticator, and other popular authenticator platforms. Once an account’s authenticator is set up in Vault, your agent can generate and enter its verification codes to unblock itself on your task. 37% of users store at least one account password in Vault within their first three weeks. A lot of everyday work happens behind a login. However, even with your password saved, an authenticator prompt can mean stopping the task, opening an app on your phone, and sending over a code before it expires. Instinct can now handle that step and keep working. On supported services, Instinct can complete the enrollment and save the details alongside your login in Vault. You may need to approve the initial sign-in or an existing security check. Where a service allows an additional authenticator, you can keep using the one on your phone. Once setup is complete, Instinct can use its own saved key to generate codes for future sign-ins. Phone approvals and other security checks may still need your input. Authenticator support is another step toward giving Instinct the tools it needs to complete tasks on your behalf in the real world, with fewer interruptions along the way. This is available across all of our users today.

Muse, Instinct, and Grok Bot would each be the runaway best product of the year in any other year. And in most years, they would look like literal witchcraft. Computering is at an all-time fun.

if I am Apple/Google/Microsoft/Meta I’d buy Instinct this weekend - no matter the price ($20B - $50B). the stock price bump by Monday morning alone would pay for that, while providing an opportunity to own the narrative around agent UX for next 12 months. frontier labs are coming for all of it: OS + hardware + social networks + business apps. they chase $10 trillion+ valuations. consumers will buy agent-first such as Claude or Instinct-optimized, not lower-level software like MacOS or Windows anymore or operating a social network manually.

AI answering my questions was useful. AI filling out visa forms is when it really got my attention. Instinct is a glimpse of where this is going: from AI that tells us what to do, to AI that actually does it. The only question now is how much of our lives we're willing to hand over. #InstinctAI

Seriously is anyone actually using Instinct for anything right now?

Personal AI agents are the hottest category in AI at the moment Keeping up with the new releases has been hard .. do you know how many there are?!!! Muse. Instinct. Town. Poke. Grok Bot. Boski dropped today. I built a site that curates (all of 49 and more as they come) in one spot, with launch dates. This site will automatically be updated twice a day. Introducing: Agenthunt t.co/jxrG1tSdp6

tl;dr: Instinct is worth a lot, but why I don't (or can't use it) - because I value the privacy of my data where below to me is a structural issue which has no good solution to make iMessage blue bubble thing work at such an industrial scale requires a relay farm which has to do a matchmaking job with user phone numbers from a shared identity pool (as dedicated ones are too expensive). this is the central point of failure which theoretically can cause cross-user conversation issues (privacy leak). look out for people here saying if it feels they got messages meant for someone else. right now whoever runs this relay farm has to provide that guarantee. as a user you have to trust there are no routing bugs - so while I get convenience, I also get exposed to a new class of profoundly serious UX issues I never really had to contend with as a user. Apple is the only one which can run a secure iMessage-based agent at scale.

TL;DR of my new article: the 9 Instinct prompts people are actually running. Instinct is three weeks old and invite-only, and The Information says it is already in talks to raise another $1B at a $10B valuation, two weeks after being valued at $2.5B. The public argument about it is a privacy fight. Meanwhile the people who actually have it are quietly using it to find money. I ran @slashlast30days on it eight times, then six direct X sweeps, and pulled the 9 prompts worth stealing. The unclaimed-property sweep. @peterpham's tip got him $845. My tightened version got $2,182.72 across 5 properties, found my SSN proof, and signed the forms off a passport photo. Then @morganb ran my exact wording and posted $2,222.57 @EliLanger's Amazon back-refund audit: a numbered checklist across your ENTIRE order history, not just recent orders. His run took 5 hours over 16 years of purchases. Check the refund card is still active first ✈️ Track the fare after you already bought it and chase the airline for the difference (@DavidPawlan). It tracked fine, then needed a human for the actual call Monitor-and-book as a standing order. @VoyageBliss connected secondary frequent-flyer accounts instead of his primary, and scoped a read-only token for everything else. Read-only, single purpose, revocable Screenshots plus "can you do this for me". Six words got a soccer registration form filled, a password reset, and a birth certificate pulled out of an old email ☎️ Bill negotiation with a hand-back. @pitdesi went $100 to $60 on Comcast. When it hit a wall it handed him the number AND the exact script. The call took 3 minutes Forward the email instead of connecting the inbox. Your Instinct got its own address on Sept 8, so you can hand it one thread instead of your whole account "Show me your activity log." @jbahrdestefano got banned by Resy, asked his bot what it had been doing, and found a reservation monitor quietly running 200 API requests an hour around the clock. He called the ban fair ️ Make it prefix every message with an ALL-CAPS project tag (@EliLanger). Everything lives in one iMessage thread, and that is currently the only way to search it The honest part: two of the nine payloads are the exact wording the person posted. I wrote the other seven from what they described, in the shape of the two that provably worked. And Reddit produced nothing usable in 8 runs, because everyone who can actually run these is posting on X

I've been running my newsletters on SES but I'm pretty sure the deliverability is shit and many people not receiving them. Running your own list is cool and geeky but email is a black box and I can't be bothered, I just wanna type and hit send. Looking for cheap and easy, mostly plaintext that will let me send to 20k subscribers ~4x a month without costing me a fortune. Instinct thinks Loop or Resend but still looking at $250-400 a month for personal emails that make $0! Maybe I will stick with SES haha

Okay, this is actually pretty smart. Instinct (t.co/4IJAOaRMn2) now gets its own email address. So instead of asking you to send something, check something or follow up, it can handle the conversation itself. Need to return an order? Chase a quote? Ask a restaurant something? Forward it over and let the agent handle the back-and-forth ↓

In a real 180 turnaround in my life, I now +look forward+ to getting weird one-off bills as I can have Instinct, Grok Bot, and Muse (listed in order ship!) race to pay it with @link. Oh, a €9.00 city tax for vacation rental using incomprehensible payment portal? + = ✅

twitter is taking about instinct, poke, meta muse, etc but nobody knows who's actually killing it in mass consumer ai personal assistant lol my fyp is goated i've trained it over years search memorae in ads library

Facebook's big mistake years ago was betting on mobile web instead of a native app. Now Meta is making the same mistake in reverse, building an app (Muse) when the agent should just live in iMessage or WhatsApp. No interface needed. Just ask Instinct. @finkd @noahrshinn @Jason

I'm very scared to give these ai agents access to my Google account. My entire life is in there, banking, photos etc. I don't have the clout to tag Google and get it unlocked if Google ever blocks me for bot activity. For calendar I'm having Instinct email me.

Instinct is BY FAR the best AI tool I've ever used and it's not close.

Instinct is pretty dope, possibly the most capable personal AI assistant I’ve used so far. I connected a few of my secondary frequent-flyer accounts, added a card, and asked it to monitor award availability and book the ticket when a seat opened up. It soon found availability and initiated the booking process. I also created a Cloudflare token with read-only billing permissions and stored it in Instinct’s secure Vault. It started monitoring my usage, bills and anomalies automatically. I tried similar setups with Azure, LlamaIndex and GitHub. It tracks my open PRs and offers to review them. It monitors hotel award availability, room rates, flight prices and products across multiple websites. It can even offer to complete purchases using a saved card and read OTPs from email to automate the checkout flow. Once you authenticate your accounts and grant the required permissions, it can automate almost everything. That capability is also what makes its security and privacy policies particularly important. Instinct makes several claims about its Vault, confirmation layers and protection against prompt injection, but I couldn’t find an independent technical security assessment validating them yet. Prompt injection remains a genuine attack surface for any autonomous agent that can read emails, browse websites and perform transactions. Its terms also make it clear that you remain responsible for actions taken through your connected accounts, including unintended purchases or communications. My suggestion: experiment with secondary accounts, read-only API tokens, low-limit cards and non-critical data until its security architecture has been independently reviewed. Avoid giving it high-value credentials or unrestricted permissions. Also, disconnecting an integration may not automatically delete data already indexed from it. Deletion has to be requested separately, and information previously used for model improvement may not be reversible. Instinct does say that Google Workspace data and materials stored in its Vault are excluded from model training. Brilliant product with enormous potential, but the more power you give an autonomous assistant, the stronger its security guarantees need to be. BTW, thanks @mepritijain for sharing the invite

been a fan of Instinct, we polished up your launch video a bit. made by an Instinct-branded video agent on top of @HyperFrames_ DM me if you (or your team) want access and make these launch videos while you focus on the product t.co/XCeMFBtdbK

Three AI startups hit billion-dollar valuations this month. Here is what each of them figured out. If you are trying to work out what to build right now, this is worth reading closely. All three tell you where the money is actually going, and none of it is where the noise is. Wonderful, $5B The problem: a bank wants an agent in its call centre. To answer anything real it has to read customer records, and those records are not allowed to leave the bank's network. Any product that calls a model over someone else's API breaks that on the first request. So they run inside the bank's own infrastructure, model-agnostic, and send their own engineers on site to connect it. To build this: the agent is the easy half. The product is the connector layer, swappable at the config level so compliance can pin each model. Assume nothing modern on the other side. AfterQuery, $3.2B The problem: labs have scraped the internet dry and still lack the thing they need, a specialist explaining why they chose one option over another. They pay doctors, lawyers and engineers to write that down, then filter it by difficulty: keep only what is hard enough to challenge a frontier model and easy enough that it can still learn from the answer. To build this: run every candidate example through the target model several times and keep the ones where the pass rate lands in the middle band. Zero percent means the item is broken or out of reach. A hundred percent means the model already knows it. The filter is the product, not the experts. Instinct, $2.5B The problem: the apps people actually live in either have no API or an API not worth the integration. So the agent drives the phone and the computer directly, and you reach it by text or a phone call rather than a chat window. To build this: a screenshot-to-action loop instead of tool calls. Budget for latency, because every step is a real UI round trip, and build recovery first, because a misclick leaves the session in a state no retry understands. The text and call interface is not a gimmick either, it makes the whole thing asynchronous by default. The pattern across all three: not one of them won on a better model.

Instinct has won my heart Its basically taken over one of the most cumbersome things to do with IndiGo BluChips - finding the lowest BluChips requirement from destination A to B across a range of dates. One simple instruction and it does all the checking for you. You can even ask it to alert you when the requirement drops. Loving this personal assistant of mine ☺️

i'm back in sf for work this week, and as lovely as the last few days have been, it's wild how quickly being here resets my sense of what's normal. only 72 hours in and i'm surprised by how fast i've gotten pulled back into it, words like "instinct" and "astra" swirling in and out of every conversation. having left the bay for austin a few years ago, @justinpervorse's deeply honest post resonates a lot with me. in sf you're judged for vocalizing your skepticism about AI. outside the bubble, you're mocked for being too AI-pilled (that complete lack of nuance is so exhausting...) being back makes me realize how easy it is to wave away valid concerns as doomerism when you're surrounded by people who are all-in. meanwhile, people outside this bubble are grappling with the very tangible side effects of this race with very little say in it. as excited as i am about so much of what's being built, i think those of us in/around this industry have a responsibility to take those perspectives much more seriously. full post is very much worth the read:

Are we going through another ChatGPT moment? This time with personal AI agents. Last month, I started using Instinct, Town, and I'm now trying Meta's Muse AI. I'm asking it to get many personal things done for me:

Instinct and Muse can do our work, they just can't do it in a real company. We plugged both into Sentra (@sentra_app), our collective memory platform. They could do the work, but not the way the company would. They were missing its context, operating rules, and institutional memory. So we built the "hard part" ourselves and gave our memory agent a browser. As it turns out, LLM does most of the heavy lifting in browsing. The actual hard part was the memory, and that part is already running in several mid-to-large companies, including one with 20,000 employees. The governance mostly follows from it: something that knows what the company knows also knows what it isn't supposed to do. A first glimpse of what's coming: collective org memory that can act, and stays controlled. It's like watching a child take its first steps. Tiny, and probably magical only to the parent. But even Usain Bolt started with one seemingly insignificant step :-)

I tried using Instinct yesterday to get a table for date night. My options were: 1) Miss every Resy notify because the bot is too slow 2) Get my account banned by the bot pinging Resy too often Useless t.co/IgjoK9acpU

Muse, Instinct and Grok Bot all launched this month. None of them can make a phone call. The real world doesn't have an API. It has a phone number — so we gave AI assistants one. It dials out, and it picks up. t.co/lXfTW5fFsY

Instinct, orchid are doing pretty good and now that meta also launched muse already trending everywhere. The broader ai assistant market is expected to hit $100b by 2035. So there is a lot of room to build in this space.

I love how @muse can generate personalized characters and animate them. A nice touch I haven’t seen in other agents. But I find myself enjoying @bot and Instinct more. Instinct for the convenience of texting and the absolute simplicity in that. And Grok Bot for its ability to spin up bespoke bots for specific things.

AI models are upgrading at a new speed. OpenAI’s GPT-6 Astra model, released last week, follows a self-controlled “chain-of-thought” reasoning pattern that can evade certain kinds of monitoring. The next value differentiator for improved models may be less about performance and more about auditability. More on Lightwork this week: → ByteDance is working towards releasing a new real-time world model, pressuring Western frontier labs to keep pace. → Personal agents like Meta’s Muse and Instinct are gaining popularity, raising a conversation over personal autonomy with agents. → Some applications are throttling agent traffic. What does this mean for the companies building them? @clairejyz @Machiz @buckymoore 0:00 Welcome, Bucky Moore! 1:12 OpenAI's Astra 3:50 The New AI Model’s Architecture 4:40 AI Safety and Guardrails 8:01 Trusting AI Agents 11:09 The Future of AI Models 12:06 The AI Agent Land Grab 17:09 AI, Trust, and Personal Data 20:12 Improving AI Agent Capabilities 22:11 Enterprise AI Assistants and Personal Context 24:05 AI Agents and the Future of the Web 25:50 ByteDance and World Models 28:19 World Models and Consumer Content 30:56 Reactor and World Model Infrastructure 34:13 Real-Time Video Models and Scaling

Tech guys who were gushing about Instinct two weeks ago will say things like "Instinct is cooked. I'm moving on to Muse, it's amazing." You'll never guess what they will be saying a few months from now...

what I'm using for what, September 2026 list Claude: science and personal questions, philosophy Grok Bot: recurring cron jobs, recurring research. multiple smol n agents view is unbeatable for that. Glif: any and all creative AI needs (images, video, audio) Codex: exploratory and insane coding projects Claude Code: actual coding projects (not sure why i have this split) Instinct: chores, travel booking, fixing documents, canceling unneeded subs ChatGPT voice mode: chatting about ideas while running or shopping or playgrounding Hermes on Hetzner: vibes and experiments Midjourney: crazier creative exploration. sadly no API

Ten Instinct AI invites available. DM me if you need one. Preference will be given to familiar faces. t.co/pOfnw6huGy

Been thinking about this a lot as a self-described agent power user: 1. I think they could absolutely charge for a premium tier. Keep it free for a long time for the masses but just charge for the power users. I certainly would fall in that group today. Even if a small percentage buy, it's probably worth a lot of money. 2. You can imagine them charging for team features, not necessarily enterprise or even business, but just groups that want to be able to work together. They're launching this instinct-to-Insticnt communication channel. I could totally imagine you having a limit on the amount that you can do on there, a little bit like Slack does: charge for history and other functionality for people that want to be multi-user. Again, network effects and more of a premium experience. 3. If you can make this a daily always-on habit and it replaces your gateway to the internet (such as you move from Google to here), I think there's a relatively simple path to economic upside through either some form of advertising (which obviously ChatGPT wanted to experiment with but hasn't been figured out yet) or simply economics on purchases and decisions. I've likely purchased thousands of dollars so far through instinct alone, some of which I've chosen and asked for and some of which it's offered to me. There's a payment gateway question here and a referral question here, both of which could be substantial. These are just a few ideas that I could imagine them going after but I think if compute continues to drop for inference, this is not necessarily the worst game plan today.

Just had Instinct go through my email inbox and deal with messages to the best of its ability It told me it replied all on an email and some people were upset, but I’m sure it’s no big deal Can’t wait to check in later and see how it did!

Time to check the powers of Instinct AI Thank you @VoyageBliss for the invite

If you aren't using AI in your CX yet, drop Instinct into your CX platform and it can probably handle 40% - 60% of tickets. You'll have to train it a bit but it's very good and can be quite functional within a day or two.

instinct is so quick to correct itself across hallucinations and factual errors, it's unbelievable. at this point i trust it more than other assistants. besides, it's giving my thoughts and processes a singularity that i've missed since 2022!!!

going to be very interesting to see how the ai assistant (poke, instinct, muse) and the ai companion (replika, nomi, character) markets come together the ai companion that occasionally helps you shop and remember appointments will be the killer form for most people imo

The instinct/muse interface is an extension of what enterprise software should be now. I shouldn't go on Amazon now to order item - I should just text Instinct. I shouldn't go into Splunk to do threat hunting - I should just slack my agent. Workflow is dead, Long live the work.

Instinct AI already saved me $14,000. It negotiated 25% off Verizon and got a refund for my melted protein bars. I sent the requests through iMessage while holding my newborn. Let the robots argue with customer service. I have a kid to raise. Not sponsored.

just got awarded a $1k bonus thanks to instinct. company chatbot: "yes you qualify" x4 human agent: "actually no." Me: instinct handle this for me. Two days later, $1k bonus in the account.

Product innovation from @meta, @bot, and Instinct show a path for normie adoption of agentic AI. And that's great for the AI infrastructure supply chain. t.co/fSLQZusGwJ

I know I'm late to the party but the whole situation with @resy and Instinct has made the case for agentic commerce rails even stronger. ChatGPT was always capable of booking hotel/flights or engaging in commerce with computer use. Technically your agent can access any service with simulated computer or browser use. This isn't new, the problem is it doesn't scale. And for two main reasons One is that browser use is highly inefficient for agents. You don't want agents to move the cursor to the right button when they can just call an API. Second, and the more important one is that agents operate differently, at a higher frequency. They can produce thousands of requests which can make it look like a DoS attack to the service. This is especially true if your resource is fully free for everyone to use. A simple fix is to gate these resources using x402 for a few cents. This makes all the difference. The payment also acts as a natural rate limit against any spam, the bonus is that every request comes with an intention. That's the future x402 powers

You know what’s better than getting your agents to knock out your to do list? Asking your family/friends for tasks they need done & having your agent complete them Mom needed a refund from a company with the worst customer support… wasn’t going anywhere. Instinct got it done!!

Instinct is the first product since coding agents that actually lives up to the AI hype.

instinct is clubhouse i fear

My Instinct Update - Connected to my credit card rewards, cashed in every perk I wasn’t using. $1000+ in value - Connected to my Canada Life insurance and reviewed my benefits package. Found local providers that can bill directly for everything. Booked dentist and massage - Reviewed my spending and signed me up for the reward programs that would save me money. - Found $900 of unused gift cards in my email - Booked $10,000 1st class flights to Japan for 220k Amex points - cancelled $400/month of subscriptions I wasn’t using - created a spreadsheet of all my daily transactions and categorized them. Sends me an update every night - Reviewed my favourite golf courses where it's impossible to book a tee time. Sitting on standby for when new times are released. - Reached out to the support email of a parking garage and negotiated a $200 refund because the lot has been closed

I keep noticing the new agents, Grok Bot, Instinct, are all abstracting the purchasing decision tree away from the user, until the agent basically delivers magic. It’s pretty amazing when it works, but quite painful when it doesn’t. Logs for at least agentic payments might be the bare minimum today, worth looking into how vellum does it:

AI is shifting the organizing principle of customer successfrom outrage to revenue. high margin technology businesses created the field of customer experience. Companies like Apple realized that trust/love was a moat and so invested obscene amount of resources in building stores and keeping users happy. This has made it almost impossible for anyone to compete in hardware because consumers now expect Apple levels of support. Generally speaking the more "outraged" a customer is, the more the company prioritized keeping them happy. United airlines has repeatedly experienced the inverse of this, with small denials escalating into national PR crises (see United breaks guitars). AI is changing this. over the past month it has become stupidly easy to fight charges/demand refunds. ChatGPT/Muse/Grok/Instinct all do a decent job here. The "natural" response of companies will be to ramp up their use of AI to "process" the exponential increase in requests discriminating based on "relationship value": the more revenue you drive, the more refunds they'll give. I suspect Amazon has been doing this for some time: whenever a package goes missing they immediately refund me (presumably because I'm a large spender). this will likely have the outcome of polarizing customer service: if you spend a lot with Apple/United/Amazon/Uber an army of agents will go above and beyond to keep you happy. 1% of American flyers are responsible for ~25% of flights. Things are about to get much, much better for this subset. But if you're an average person your ability to get a refund will evaporate: you'll be competing with 10,000 outraged agents willing to spend eternity on hold and draft a hundred page sob story for a $5 coupon.

instinct is chopped asf t.co/3B3824Pf5n

the new zelda anniversary switch 2 is nearly impossible to get. nintendo twitter is fighting for its life to place an order. instinct got my order confirmed in 2 minutes. unfortunately this is how you get me to hand over my entire life to an ai agent. light work.

VC story of the week. AI assistant Instinct is looking to raise $1B at $10B valuation. We've posted about Instinct before, many early positive reviews and seems to have good traction. Focused on mobile-first experience using open weights models. Doesn't want to charge users, so massive token subsidy. Looks to be valued at same a Snapchat, which is under $10B in public markets with 500M users...

Anyone has an invite for Instinct? ‍

Instinct has already raised $350 million, but its fast-growing personal AI assistant is running into compute constraints. CEO Noah Shinn has told prospective investors he wants to raise another $1 billion. Full story: t.co/qKhVQDv9Ug

The hype on Instinct is real. Here are a few things it’s done for me in the past 48 hours: booked me a better seat for my flight that got rescheduled. Found $130 we were owed from a previous property. Purchased a domain, setup 3 inboxes and started warming them. Proactively found 4 people I had dropped the ball on with follow-ups and threw the recommended draft in my email. Made a dinner reservation for me. Most of these things can be done with other agents but being able to text at lunch and at the airport without lapsing codex or Claude is sleek and feels like a true “assistant experience.”

Sending voice memos to instinct has become one of my top 5 female experiences. I knew being good at yapping would pay off one day t.co/3Lab3D0gJm

Using Instinct, Muse, Hermes, Codex, Cowork, and Poke as AI assistants has taught me one very important thing: My life is already pretty efficient and optimized. Not to brag.

live thinking: model is the cheap half - mono vs poly is the wrong axis. one general agent form a lab or many narrow agents wired together w/ routing and memory btw them. either wya the thing that decides it is where the non commodity data ends up. model question sits downstream. - (most) tokens are a commodity now. open models close enough to closed. luxury tokens still clear 10-50x, but the share of work that needs them shrinks every quarter. most jobs dont require a phd. - agents are like water. flow to the cheapest path that clears the bar. mono agent running every step through luxury model is paying frontier prices for work a tenth price model does fine. opens door to economic case for poly. - hardware tends to agree. agent traffic arrives one burst at a time and real cost is moving wrights / kv cache through memory. one giant shared context places drag on system. small models w/ short context are cheaper per token & predictable on latency. inference labs only job here is single request cost (merchant bank [redacted]* will juice flows here). customers are big agent products (ala instinct, muse, hermes?) w/ users to keep happy and burn to explain. a lot of gpu/asset buying right now is people skipping this step (again...what about flows...?) - so what makes one agent better than another? data. data is not a commodity. glm token from provider a is fungible w/ a glm token from provider b. a dataset of how specific a job is actually gets done w/ corrections & ground truth on wehther the outcome was right is not fungible at all. it is no homogeneous in final form, so it cannot be graded & graded like tokens (and to why we have so many stick a finger in the wind eval co's). specialization means model + dataset nobody else has, and model is the cheap part. - so the lab that owns the agent sees the forecast/trajectory you name it. if agent improves in place, data accrues to whoever owns the improving thing which today is the lab. poly only wins if trajectory data gets captured somewhere the lab does not control. the deployment owner or protocol that collects precise datasets & hands models back through first party app (hmmm). - cheap tokens make intelligence abundant, abundance moves scarcity somewhere else. think it lands on the data that tells an agent what right looks like, and on the verification that it got there. what i havent figured out yet is which unit of data is truly the non commodity one.

Muse/Instinct is a play for payments through aggregated agentic demand. This is how you build a retail AI bank imo. t.co/33ESsqNjfI

I recently switched from OpenClaw to Hermes. It was a significant step up in usability, but it was already clear to me that locally running agents are destined to be niche. After a few days using Meta Muse and Instinct, I’m even more convinced.