We probably need to all update our sense of what’s coming in terms of agentic workloads with the combination of agent swarms, better computer use, the next wave of APIs and MCPs coming online, and new form factors like Muse or Instinct, vertical enterprise agents, background workflow agents, and other products that are emerging right now. We’re going to throw agents at vastly more tasks in our professional and personal lives than we had initially imagined. The kind of information that agents will go out and find for us and do work for us in the background is going to be 100X more volume than what we can imagine previously prompting in a single session. You’re going to have agents go out and surgically recruit talent for you 24/7, look for every signal in your customer’s business for when to pitch them, process every single transcript and conversation for product insights, review every line of code written for security issues and bugs, brute force test your systems for issues, and 100s of other tasks. We’re probably 1% of the way into the shape of what all these agents look like, where they get deployed from, how they get managed, how we budget for them, and so on. But it’s inevitably going to happen at a scale that wouldn’t have been fathomable before.

This is the kind of AI that actually feels useful. I asked Instinct to organize a Friday night with friends. It coordinated with everyone’s Instinct, figured out food + drinks, picked a movie, synced everyone’s timing, placed the order and added it to the calendar. No endless group chat. No “what time works for you?” No chasing people for answers. Just: “Lock it in.” And it’s done. AI assistants are finally starting to coordinate real life, not just answer questions.

ok this guy actually cooked For anyone who missed it, Noah Shinn built an agent called Instinct. No app or interface. You just text it like a normal contact. He just shipped an update and now you can literally tell your agent to sort something out with your friends and book everything. It messages their agents, collects everyone's availability, finds a slot that works for all of them, and comes back with the place and the price per person. You say book it and it's done. It also has its own email address now, so it signs up for things and chases confirmations without any of it touching your inbox. I am genuinely impressed.

Yet to see an example of something grok bot, instinct, muse etc can do that codex or claude can’t do with the same prompt

Source: AI assistant Instinct seeks to raise $1B at a $10B valuation, up from $2.25B less than a month ago; Sequoia and Benchmark are in talks to lead the round (@pauvalida / The Information) (Visit Techmeme dot com for the link and full context!)

Meeting with NVIDIA Sr. Director went great Key takeaways: - the industry has gotten the message. Everyone is running to support - we are still “very early”. Only 2% penetration globally for Ai users. “Instinct and other personal assistants are exploding usage” - NVIDIA leadership is working 6-7 days a week right now unlocking more capacity and solving key problems - energy needs exploding not just for data centers, but for all domestic manufacturing including manufacturing racks, components, tools, and fabs - re-industrializing America is well underway - there’s still so much work just to enable Vera Rubin, let alone Vera Rubin ultra and later, Feynman. The amount of strategic and “insanely hard” work to enable these ramps is not to be taken for granted - benchmarks like what we saw with Jalapeño from OpenAI recently are only scratching the surface of evaluating alternative solutions to NVIDIA. Need to evaluate at the system (rack, pod, cluster, multi cluster) scale - it’s not just about working insanely hard at NVIDIA, the team really feels they’re contributing to history. Which seems to be the primary motivator for them since many in the organization no longer need to work for money - there is no “biggest bottleneck”, the bottleneck is shifting and present at multiple places at once in time then appearing somewhere else. It’s whack-a-mole - Physical AI is next, and we have not even begun with it hardly at all. But Physical AI will take many multiples the amount of compute that we currently have

Inside Instinct AI, the $10B invite only personal agent app. Is it worth trying? t.co/1KTSf39Ofn

How big is WebMCP going to be? Instinct, Muse, Town, Grok Bot, and most AI assistants are converging to the same architecture - a headless browser + a credentials vault to inject passwords, credit cards, and personal information. So what's the best way for that browser to actually use websites? WebMCP.

I’m hiring a Head of Technical Investing to lead the work on my personal investments. My work at Abundance is shaping my view of how long-running agents, continuous improvement, and proprietary data will change entire industries. This perspective has led me to invest in Cognition, SSI, Instinct, Forus, and Trajectory, among others. I’m looking for someone to take ownership of this function: sourcing and evaluating opportunities, investigating the technology and business independently, and bringing me well-supported investment recommendations. You should be able to go extremely deep technically while keeping in mind the commercial implications. Former founders are especially welcome. Investing experience is also valuable. You don’t need to arrive having done every part of the job. This is a 12- to 18-month tour of duty for someone thinking seriously about what to build next. You’ll spend that time evaluating companies, technologies, and markets in depth, sharpening your judgment about what you want to build or join afterward. I’ll support you in that next step. This will be intense, and that should be part of the appeal. Send me the following: - One or two areas where you have exceptionally deep expertise. These can be in any field, however niche—I want to understand your strengths and where you’ve had substantial firsthand exposure. - An investment idea you’ve researched deeply: what others are missing and the evidence behind your view. Email: jobs@apoorvamehta.com

Unexpected benefit of using Instinct: being able to use the internet over WhatsApp/iMessage when your flight wifi is spotty

So Meta $META has or will have: - Top tier model (muse and watermelon upcoming, will lead to in house model usage, savings will flow) - Muse agent - overtook instinct chatter - Massive distribution advantage - Massive amount of compute - Data center land and power secured - at least 5GW in Louisiana alone. - In house chips - AMD chips, partnership, and options - if MTIA doesn't work out, ramping up AMD procurement is a decent fall back plan - Core business ripe for AI deployment and integration.

bro what is this instinct ai agent it made director stamps on my name and stamped the agreements and mailed to the party, gave me follow up, helped me made an entire diet plan for the client via voice notes on pfc backend app and ordered protein for my mom just via voice note

I emailed Instinct asking for comment regarding the conflict in their service, terms of service, and privacy policy. Read the full email here - t.co/gHbSjLXN5t I'll update if they respond. t.co/5QIVmAUgfN

don’t worry bro my grok bot is on it. yeah no problem i gave it to my muse. trust me bro my instinct can do it.

if anyone still needs an Instinct invite code, comment below and I'll DM one to you mine all expire today

one week ago today, I decided to create a benchmark that would compare every AI assistant out there it started with 3: Instinct, Grok Bot, & Poke today, we're up to 108 - 70K+ website page views - 80+ organic assistant submissions - 235 tests ran across 23 agents (so far) - listed under X trending A lot more work ahead, but damn has this been fun

Given the hype around Instinct, Muse and consumer AI, I’m resurfacing something I wrote 6 months ago. The key to mass consumer adoption is removing the prompts. They should run like auto-pilots, only seeking confirmation at important junctures. The product that wins will remove the need to prompt altogether.

Okay I have to say I haven't had much use for Instinct whatsoever. It's booked me two little van transfers to and from an airport, which was fine, but now I have all of these dining reservations I'm making for a trip for myself. I want to contact the restaurant and tell them the occasion, and email another restaurant and another cocktail bar about stuff. I just told Instinct and it did it instantly so I see the use case now for some of these things now for sure

3 months ago agentic commerce didn’t exist on my book as a consumer Today if a business has friction to shop using Instinct or Muse I rather buy elsewhere - the experience is just so good Imagine where another 3 months will get us or 3 years

With a single message and zero previous context, Instinct fetched the PNR from my email, checked me in, and sent me the boarding passes

hey instinct how are prediction markets going

Instinct is freaking amazing, stop coping

Hot take: Instinct can’t escape its enormous valuation because: 1) it’s not hard to build thanks to AI. I am building a similar tool, it watches their twitter and queues up new features to stay at parity 2) there is maybe no moat here. If they had years of head start maybe but @bot and @Muse are already here and already great. 3) founders should think deeply about what this means for their business

FREE IDEA: @muse or @bot or Instinct should just sponsor the world’s most ambitious hackathon for using AI agents A huge opportunity in front of us to turn society’s notion of AI doom into into techno optimism Ideas? Encourage people to use personal AI agents creatively and ambitiously for net good > Use an AI agent to help bring a park to your neighborhood > to report potholes > to save money by shopping around a deal on car insurance > to find your dream wedding venue > to solve a cold case > to find a cure or meds for a rare disease Endless possibilities Whoever gets to do the most ambitious AI task (s) .. wins special prizes (across categories) A million agents doing a million positive things can change the world The user sentiment is brewing and ready. Just need a worldwide movement around this

I disagree. Possibly my biggest judgment error was assuming value in AI would accrue to Google (vs OpenAI/Anthropic) bc they had the distro, balance sheet, and headstart. Seeing something similar happen with Instinct v Muse…distribution isn’t the same edge it used to be. t.co/zvHKaemSO7

someone invite me to Instinct pls ty

I’m on Instinct guys. Lemme know who needs an invite. I have like 15 of them and they expire on Thu

TL;DR of the new /last30days podcast with @FarzaTV, founder of @heyclicky (I'm a paying customer and I begged him to let me invest). The thesis: $ , , ' . Grok Bot is shipping a feature roughly every other day. My hacks article hit 1.2M views, @elonmusk reposted me twice, and @Scobleizer endorsement was " ' " ‍ Many . Thought it was a prank until three separate people showed me screenshots Instinct went from unknown to $. . Index led the $250M Series B. The pitch: no app, no interface, you text it and it drives a phone and a browser like a person would I tried to buy one vinyl record. Grok Bot failed. Handed it to Instinct. Failed. Stripe reached out and my bot sent them the receipts. Farza on the Grok Bot hype: " . ' ." Will he keep using it? " ." Is it inspiring? "100%" ' . The entire HeyClicky product except the landing page is generated code and " ." His favorite flow: take something he loves in his own product, paste it into Claude Code, and say brainstorm with me about why I like this ️ His secret internal tool is : every transcript, message, and document the company produces gets dumped in, and an agent maintains a living Wikipedia out of it. " ' " ✈️ The OG team: a Peruvian contributor's fork of Clicky got 12 views on Twitter. Farza FaceTimed him in his car in Texas: " ?" He said yes Dad drove a taxi in New York, made pizza in Miami, then sold Nissans starting in 1999. He resigned two weeks ago. At 13 his friend was spinning a sign for $12 an hour and Farza wanted in. Dad, at the dinner table: "' . ' . ' ." So he sold a $40 Calvin Klein shirt from the donation pile on eBay instead That turned into six years of selling blank media and clothes online. Thousands of dollars a day sometimes, all of it going back into cameras and costumes for the YouTube channel. , . " " Every night at dinner: " ?" "$2,200." " ?" "$3,000." " ?" : , , '. Everything else was allowed, including Mom driving him 90 minutes to buy inventory from a guy he met online How he got to SF: his League of Legends computer vision model went viral in 2017 and every company in tech hit his inbox. One guy ran a company tied to " , ' . 22." Got in at 24 with Zip School, which pivoted into Buildspace On the Pakistani flag in his bio for a decade: growing up in Florida he never saw anyone who looked like him building things. "' " On the Steve Jobs comparisons: "' ." Then the real point: in 2018 he and his friends idolized Collison and Chesky. In 2026 he asked his brothers and cousins who they look up to in tech and got " " " ' . ' ." He told his YC batch the story of how his teenage YouTube videos got him in trouble, they all freaked out, and he refused to tell it again on the pod : , , ' . " , . , ." He's asking about the next video before the current one has finished uploading. " ' ' " ✍️ The creative hack: , , on his own free app (t.co/yHoTHP05nC). The hooks fall out of the writing The advice he gave his whole YC batch: open Reels, record 30 seconds unedited about what you're doing, post, repeat for 30 days. . They hired agencies instead. " ' " On the Buildspace shutdown: " , ." A single bad comment would live in his head for weeks. Now he won't answer questions about competitors or milestones at all. "' " " , ." He'd rather be on this podcast at 70 than stop at 35 because someone handed him a billion dollars " : ' ." His regret: he could have kept Buildspace alive an hour a week from a Starbucks and it would be doing better today just for having survived. " " My version is . ' . Farza said he heard it for a decade and never understood it until he closed a company Bonus from the top of the show: Spider-Man Brand New Day crossed $900M domestic in 36 days and Tom Holland is clearing $100M on $20M upfront plus uncapped backend. Also I made my dad a Suno rap for his 79th in my own cloned voice and it is terrible. , , . That's the whole episode. Full episode below.

Instinct has become noticeably slower in its responses since I first started using it.

the biggest opportunity may not be building another personal assistant it’d already be huge if anyone outside of @Muse, @bot and instinct can stay in this space long term. they have a massive head start + almost unfair distribution advantage my observation is a lot of startups can be built around this wave too: infra, identity, payments, memory, permissions, security, distribution, and tooling for agents you don’t have to win the agent layer you can build the picks and shovels around it

There are too many AI personal assistants I can't keep up anymore @openclaw ↓ @NousResearch (Hermes) ↓ @interaction (Poke) ↓ @pallyai ↓ @bot (Grok Bot) ↓ Instinct by @noahrshinn ↓ @Muse by @Meta ↓ Persona by @zach_yadegari t.co/APA8Ms7u0N

I've been tinkering with both Hermes and Grok bot for the past few weeks. Here's a note on what each of them are good at and what might be a better fit for you. Hermes is closer to a lego block set. It takes time to set up, but everything is customisable. You can choose the model you want, the place cron jobs deliver and what it does. That can mean a steep learning curve for people who don't have the patience for it. If you are tinkering with Hermes - the fastest way to do it, is to go through their cloud deployment, use deepseek v4 flash, and plug to whatsapp/telegram/imessage. Ask the agent (via session chat on terminal in the cloud) to not send outputs that are mid-way through generating. Now you have something, that looks and functions close enough to instinct. If you change the model to Fable or Astra, the outputs are incredible. The point of a cloud instance is having a machine doing things while you are afk. Hermes comes pre-set with a bunch of tools that give it browser access. If you want to go further - load exa with $5 and set up a keenable api. I like to read niche, ancient articles and this is what I generally do with it. I type into whatsapp whatever is on my mind and it usually pulls things from arxiv/internet archive and adds it to my readwise. If you give it a Sora API key (costs ±20) it can also pull twitter feed info. Treat Hermes - as an interface between you and the web. If you want to get more sophisticated - use the cloud instance, set up webhooks and have jobs that run on specific things happening. A rather ugly example could be - tracking flight tickets to a certain region, when odds of conflict on polymarket go above a certain price. Or tracking transactions happening from a wallet. Or scanning a specific subreddit for trending content delivered to your inbox. Now onto Grok. Grok is the enterprise product that was built by start-up folks. The best use of the $300 plan is the inference you get on grok heavy - instead of the bot itself. So if you have grok heavy (the sub), and you are okay with it - you can use the Hermes harness with Grok API and use both the bot AND hermes. You can get grok's entry level plan but odds are high you will be out of usage before you set up your accounts and the first batch of cron jobs. Now why would you do that? (AFAIK) - Grok's marketing suggests a cloud computer, but all their agents run on the same machine. So if you have a heavy throughput workflow, you are looking at your agents lagging sooner/later. Hermes skill/mcp set up means you can swap out the model and use something else if you run out of mileage. That said - Grok is exceptionally smart with a few things Its agents chat delivery system is a lot easier to work with than the learning curve for Hermes. The cron jobs deliver where you'd expect it to. And they are clean outputs. I liked that. The bot marketplace and integrations - remind me of Apple's store. Hermes' is open, but it has a steeper learning curve for someone starting from scratch. Grok has the benefit of social proximity to a lot of SV folks talking about how they use it directly in their workflows. Its onboarding of your work tools - gmail, slack, notion etc is much more streamlined. So for someone who has 20 mins to be good to go, this will always be the preferred tool. For whatever reason - the agent is exceptionally smart with looking at emails, slack and notion and triaging what needs done. I would say its an 8/10 EA. One of the things Grok has going for it is the twitter integration. You can get tweets synced directly to a grok bot and have bookmarks saved on tools like Readwise or Cubox - which is what I like to use it for. The subscription gives you $50 of API credits on X, which is handy to set up some of these work flows. It also has mobile apps that are live and functional today. Which matters if you do not want to tinker with a messaging app. I see Hermes as a very powerful tool that needs quite a bit of tweaking. And Grok as a very useful tool that lacks quite a bit of power. And in some sense - that is the reality of it. Hermes is built by ±3000 contributors by a startup that (afaik) is ±2 years old. Grok sits within xAI and SpaceX's broader mandate. One of these are hitting way above their weight. Now - where do I see value? If I had $300 to burn without a care -and that meant my worklife got better, Grok is where my dollars would go. If I had $20 to burn, and I wanted just as much value but I do not mind spending a few hours tweaking and figuring how cron jobs work - then Nous is where the $ will go. - Grok is like iOS. Closed ecosystem with taste. To be precise, it feels like an iPad. Highly functional but restrictive. - Hermes feels like a 2018 era thinkpad running linux because it is powerful and works well but has some hiccups along the way. The keyboard (or skills/mcp and modularity of model switch ups) and the modality with which you can interact with it - feels like a step up to how we interact with the web. Can taste, modularity and functionality in the context of these harnesses intertwine? I think yes. The gap in whatever i call "taste" above will probably close in the next six months because Hermes will release its own apps. And fwiw, their cloud deployments are already far better than what Grok does. tl:dr - most of the things you see on grok vs hermes lack nuance. use claude for 90% of the things you do if you're on a budget. use hermes if you have the time to set it up. use grok, if budgets are not a constraint and you have a job that does not require heavy throughputs. P.s - I have no affiliation with either of these guys. I do spam the folks at Nous research from time to time with product feedback and that's about it.

Thought today while in the gym and using Instinct/Grok Bot/Astra a lot over past few weeks. I wonder if we keep wanting smarter models partly because, with each generation, we become less willing to guide them. Something I’m happy to write detailed instructions for today is something I’ll expect the next model to just understand. The task itself might be exactly the same, but now I want it done with less context and less involvement from me. You can get a lot to work today if you break down every step, supply the context, and correct it along the way (we’ve hit this threshold already; great for well scoped, repeated tasks). But doing that for every task and interaction is a lot of work. If I’ve explained a preference or corrected a mistake, I want it to remember and understand when that lesson applies elsewhere. I shouldn’t have to anticipate every variation and write another rule for it. That’s the push-pull I’m thinking about. As the models improve, we give them less to work with and expect them to infer more. Maybe we hit a threshold where they’re reliably good enough at this. But especially for people using agents across lots of different tasks and agents, I suspect the desire to explain less keeps us wanting smarter models for longer than the difficulty of any individual task would suggest.

LIVE: → YC Demo Day Recap → Thoughts on @Muse, Instinct, AI Assistant Wars → Reactions to Astra AGI Claims → Reviewing Startups Also on twitch . tv /brycent t.co/CfqKNcYSIQ

i now email my co-founder's instinct directly. @VarunKhiatani this is about to get so fun

Just came across this Bezos banger: "If you have a business model that relies on customers being misinformed, you better start working on changing your business model." Who suffers the most once everyone has an agent? - Premium credit cards --> high membership fees w/ "thousands in value," & the whole thing pencils bc you forget to use the credits. Agents don't forget (yes, callback to the piece I wrote on this earlier this year) - Phone + internet --> there's the price you pay & there's the price you get if you call + threaten to leave, & carriers count on most people never calling. An agent will call every time - Insurance --> renewal quotes are based on how likely you are to go get another quote. An agent will get another quote every year We're already seeing Instinct + Muse drive huge savings for consumers on 2 + 3 None of these pricing models survive once every customer has an agent that actually pays attention!

instinct has ruined emoji reactions. sad.

Not saying I agree with this POV and obviously Meta fast followed in an extremely aggressive way but if Muse ends up becoming the dominant consumer agent interface/tool after Instinct/Grok Bot/Poke tried to create the categories, wouldn't it score a win for incumbents w/ big war chests? Kinda begs the case to have some 'competitive' internal model capabilities (not necessarily THE latest frontier but ability to fine tune the latest open weight/open source) + fast ability to deploy and integrate into existing user surface areas, where incumbents have obvious distribution + platform advantages.

A few extra Instinct invites just landed in my inbox! If you’ve been waiting for one, DM me... I’ll share them on a first come, first served basis. t.co/ESY7UuUcPc

we brought the only black mirror episode with a happy ending to life: this is @ditto_dates: imagine muse or instinct, but for finding you good dates through simulations 185k college students are already on it our goal is to help people spend more time with each other than their phone no endless swiping, just real moments world's first agent that finds you love

OPEN CALL: If you are building a site or a web app using a personal AI agent like @bot, @muse or Instinct (or others) and deploying to @Netlify … reply below with your links I’m creating a mega directory and will continue to promote this week For the best 3 builds, we’ll gift you 3,000 credits

My AI tier list for $20M+ businesses will probably annoy a few people. Grok Bot and Codex: S tier. Perplexity: D tier. These rankings come from how I actually use the tools. Here’s the full breakdown: S TIER Grok Bot: The most reliable agent I’ve used for repeatable routines. When a task needs to happen again tomorrow, reliability matters more than an impressive demo. Codex: My pick for getting work done. OpenAI’s models combined with the environment to carry a task through to completion make it one of the strongest tools in my workflow. A TIER Hermes: Open source, flexible, and rich in context. You can connect your tools and choose your models. That flexibility is a big advantage, although it still comes with problems to work through. Claude: Still my favorite for design. Claude Code and the models are strong, but token efficiency keeps it out of S tier for me. Instinct: A newer agent that handles complex tasks while being easy to set up and use. That combination earns it an A. B TIER Wispr Flow: I use it every day for dictation. Useful enough to stay in my workflow, but I don’t see a huge product advantage over the alternatives. Granola: Good meeting notes. I use it daily too, but its role in my workflow is fairly narrow. C TIER OpenClaw: I used to be a big fan. Reliability has become the issue for me, and Hermes has overtaken it in my workflow. Gemini: Useful for understanding videos, design, and images. It’s catching back up, but it hasn’t earned a higher spot in my stack yet. D TIER Perplexity: I barely use it anymore. Between OpenAI, Claude, and Google’s AI search, I’m getting what I need elsewhere. The interesting part: I use two B-tier tools every day. A lower rank doesn’t automatically mean a tool should disappear from your workflow. A narrow tool that does its job can still be worth keeping. My test is simple: Does it do the work reliably? How much setup and supervision does it need? Does it earn its place alongside the tools I already use? Full ranking in the video.

One of the best things I’ve done is given my custom agent direct access to Instinct, Poke, Muse, etc I literally never use any of them directly Instinct for general requests Poke for human agents/phone calls Muse for Browser Use/payment integrations

Why does Instinct feel different? Its memory allegedly forgets 90% of you, but that’s actually the point. @anikamirzaa breaks down @ashwingop's thesis: "Everyone's question is, how do you solve memory? Most AI assistants have a database of memory: all your Granola notes, transcripts, everything." "What Instinct seems to have figured out is identifying what is and isn't important, and forgetting the unimportant stuff. That's the difference between human memory and an agent's memory built on a database." "Humans forget 90% of the things that happen to us. Just ask Maya and me. We're still good at our jobs because we remember the 10% that matters. When agents try to remember everything, they remember nothing." Anika’s question around the idea that this architecture is more secure: "I replied and said, I don't really get that. If it still remembers everything most important about me, my Social Security number for example, that's still a decent amount of information." "And he said: in a breach, it makes the radius much more confined, because it's not retaining your full email inbox. It’s meaningfully more secure."

Scoop: Instinct is in talks to raise $1B at ~10b valuation. User base has grown a lot to at least 100K+. Sequoia, Benchmark are in talks to lead.

Instinct is the best thing that could have happened to personal AI, and we're launching our own personal AI startup in 9 days. Here’s what we learned: 1/x

It feels like every week a new AI tool or model comes out. Just this last month... Grokbot, Buzz, Lindy Team mates, Instinct, Muse + a dozen others. It's so easy to go all green eyed and think... I WANT IT. But I don't have time to rebuild my whole system every day with a new tool. So I don't. Instead I look at whatever is new and ask: What parts of this could I build into my existing system? That's why I've had the same system for over eight months. And it keeps getting better because I'm not switching all the time. Choose your AI tool of choice (I choose Claude Code). And build on top of it. Rather than abandoning systems constantly.

personal AI assistants (instinct, muse etc) won't lead to to claude code like ARR growth bc few incremental users will pay up we learned this lesson many times but seem to suffer from memory loss? it's the reason OAI (or box if you prefer pre AI analogy) pivoted to enterprise

New AI assistant wristband just dropped. Imagine if Muse, Grokbot, or Instinct had a wristband that let you interact with your agent just by talking to your wrist. It can handle tasks for you, and you can activate note-taking mode by tapping it three times. Very interesting. t.co/uinF5d9gfb

yo instinct is actually kinda lit big fan of this just being in imessage where i already yap all day for these kind of task

My instinct agent already took 146 different to dos off my list. In less than one week! My 8 favorites: 1. Our kids need passports but the process felt overwhelming. I wanted the logistics handled without handing over my kids' private information. Received: A checklist of what to print, plus the appointment set and on the calendar. I kept the private forms. 2. An inbox with 13,871 messages It had 7,307 unread messages and too much old stuff to sort manually. Received: 4,898 messages all organized. 302 unread. A daily pass now keeps it from getting back there. 3. Find a rule mid-game for the softball team I coach. I needed fast answers without searching a PDF from the dugout. Received: a custom searchable rules website gives me the quick answer, with the source rule underneath. 4. My personal info was all over the internet. I wanted to know what exposed my family and what could come down. Received: My cell became removal priority one, free opt-outs started, and confirmed the kids were not publicly linked to our home or school. 5. I forgot half the things I had already done professionally. I have been featured on podcasts, panels, articles, and talks but it was scattered everywhere. Received: Twenty-six verified media moments now live in one searchable @Notion database with a Quote Bank. 6. A subscription renewed for $199 The written policy said renewals were not refundable. one warm email got the full $199 refund. Received: Total confirmed money found and saved reached $488.86. 7. I was over the same five meals The kids are picky, so random recipes were not going to fix dinner. Received: Our proven meals are in one @notion database, with a simple plan to try one nearby new food at a time. 8. Our vacation lived in too many tabs across family members. The house, island plans, activities, and countdown were scattered across the family. Received: a custom family site that keeps the trip plan, booking links, and live countdown in one shareable place. Let me know if you’d like more of these and bookmark this for when you need to give your agent more tasks.

Two companies were valued at $2.5 billion on the same day in August, and only one of them had customers Linear had $100M ARR, 40,000 paying customers and 177% net revenue retention. Instinct had launched seven days earlier. ▫️ $350M raised ▫️ Users who buy through it average $1,300+ a month ▫️ 32% already share credentials through Vault ▫️ Stripe partnership for one-tap, one-time-use cards ▫️ Invite-only while they scale compute Jason Lemkin's pushback is fair: Replit and Lovable were trivial to clone in year one and only got defensible after shipping 100x their launch version. Whether the moat holds is a venture question. Whether it saves you six hours this week has an answer today. Full playbook, setup order and invites: t.co/KhCQXRpio4 Meta is building the same thing with Muse: t.co/2S288wtmGe What I've written on the business side of AI: t.co/alAOQVA64p Are consumer agents the next platform or the next feature?

Instinct is the new Clubhouse.

there will never be one agent someone uses. MUSE requires a Facebook account; Instagram requires an Instagram account; Claude and Codex did not let you sync sessions; to see your Twitter post, you need a Twitter account and need one to use grok Grok; Instinct requires your phone number, and so on

my predictions for OpenAI dev week releases - sorting our confusing setup of codex , chatgpt work and chatgpt and might be brining in personal agent like muse and instinct - faster sol , terra and luna - optimized astra (both on inference on harness side)

Prime case of how Instinct helps my life as a ENTP founder. I just entered into a signal-less elevator and meeting client in 5 mins but I have an update in my doc. Codex and grok bot can do the same but the iMessage just makes it so easy as if you are talking to your intern. t.co/VKjgeyqnPi

Meta launched Muse. xAI/Cursor launched GrokBot. Instinct raised at a $2.5B valuation - pre-revenue. Technology is orienting towards AI-products that acts autonomously in the world on some stated end and/or behalf of humans. For sport, this means we will eventually have machines employable as data analysts, data engineers, data scientists, and eventually as scouts, in their own rights. (Not as tools.) Today, these are called “proactive agents,” but I wrote last month that framing them as actors (ala “machine actors” & “human actors”) is a sharper and truer mental model for what’s coming ahead. I also made some predictions, some of which have since resolved: 1. Software products from the major AI companies slowly then dramatically switch from the agent-model, where humans supply direction, to an actor-model, where humans in the companies themselves supply the foundation models' telos. 2. With the actor-model, a machine as an employee or contributor to significant products becomes possible. 3. The actor-model makes less significant the need for long-running and persistent agent who "don't sleep." The premise of an actor is that it can take initiative about when to participate or idle. 4. The actor-model will lead to a dramatic breakthrough in consumer artificial intelligence because it is more intelligible than the agent-model, which benefits domain experts and developers. 5. Applied artificial intelligence companies who launched and scaled human-facing products in the agent-model will pivot or be usurped by competitors who leverage the actor-model. 6. Scientific pursuit will benefit the most from the actor-model, and there will be many breakthroughs that hinge of affording machines free-will to go through and repeat the scientific process on their own accord. 7. The most useful technological implementation of the actor-model will let humans spend more time in reality. You can read the short essay here + share any feedback you have. t.co/frUKfgPPL4

instinct has initiated a new trend: vast products that live in your messaging apps—whatsapp, imessage, etc. this kind of existed in a bubble on telegram, but will almost certainly expand to broader consumer

1000% agree Multiplayer AI has so many fascinating permutations: - single vs. multi-agent/subagent - human and agent 1:1 collaboration - human / agent / human group collaboration - group human / BYO agent collaboration - hybrid agent to agent collaboration (eg Muse / Instinct / Grok Bot talking to each other without me) Agents are not humans. And it was much easier to defend putting AI behind the text box when we were in the chat era. There are far more open questions on how the interface evolves as entirely new behaviors emerge. It’s easier for me to imagine this as a new species coming online and humans learning how to communicate and adapt to them. So much to be designed, invented, discarded, and built here.

dm if you want an invite to instinct

What does Instinct AI do well? I found 6 things. 1) It uses iMessage features. I sent it a Game Pigeon cup pong game, and it played its turn immediately. 2) It understands reactions. When it asks for approval, I send a thumbs-up, and it reads that as yes. 3) It remembers. I told it about my monitor box. Later, it filled in my visa on arrival card and added the monitor as oversize luggage. 4) It contacts me first. I asked it to find a desk in Bali. The next morning, it told me that the same-day order cutoff was 12. 5) Its computer use is fast. 6) It has its own email address, so it does not fill my inbox. Each task took 10 to 15 minutes off my day. Together, those minutes add up. I do not use it for business or work, but it is perfect for personal use. Give it your haircut and restaurant bookings.

I have tried all of the popular AI assistants (ex Muse, geolock), my take: right now, Grok Bot is for 90% of people, Hermes is for 10%. Instinct is simply too scary with how assertive it is, and how little under the hood operations it exposes. Grok Bot wins in simplicity and unmatched levels of integration. It’s implementation of computer in the cloud is close to perfect. The bots are determined, but do not break things on their own. The logic of creating your teammates is easy to understand. The only thing I really do not like is the writing — no matter how hard I try to steer it in a more humane direction, Grok continues to write in some extra dry, compressed, jittery way. Hermes is endlessly customizable, let’s you change models and tools on the fly, own the deployment, modify its interface. But it does not have it own mobile app, for example, and definetely tailors for a different kind of audience. I find its own take on bots less useful. But it lets use plug in OpenAI and Grok subscriptions, or rotate between providers via Nous Research. Which means you are basically NEVER out of limits, and can use creative model pairing, like Astra orchestrator + DeepSeek executor. Now, that beind said, Codex (GPTWork) and Claude are awesome products on their own, just built from a different point of view. But the mass, non-work adoption is for bots. Will see how far people with infinite token budgets can take them.

I watched 4 different people try “text this person” as the first use case after onboarding to Instinct All non-tech people Interesting how common it is, maybe a sign of how much people want to delegate digital communication. Or an insight into an obvious sniff test on “what can personal AI do for me?” when the form factor is housed in personal messaging Either way, it never worked and the edges of personal intelligence are still jagged

My most heavily leveraged bot by my bot team: Company Brain It stands up a Karpathy / GBrain-style wiki on a Fastio workspace (company, project, or team), turns High Intelligence on, and keeps raw → wiki compiled. Self-improving knowledgebase your agents tap into shareable with agents inside Grok Bot and outside it (Claude, Codex, Hermes, Openclaw, Muse, Instinct, etc...). Requires a Fastio free-trial or subscription. t.co/VXImUncdHV

I’m breaking down my AI revenue stack live, and I want you there. You’ll see how I’m using: 1. Grok voice bot for speed-to-lead 2. Meta Muse for prospecting + follow-up drafts 3. Instinct Agent for persistent text follow-up 4. GPT-6 Astra for finding funnel leaks Free live session Wednesday at 1:30 PM CT / 11:30 AM PT. Register below

Call me insane but I resist using instinct or any other pro-active messaging assistant as a ‘body-doubling’ replacement because im scared to one shot my intrinsic motivation and ability to start tasks on my own, maybe abstracting out too much but ‍♀️

now we have to switch from instinct to muse. Exhausting. Can’t keep up with the AI assistant trends. t.co/4gXoGd2e2R

whenever i hear instinct for x, i’m like why do you need imessage for your twitter

somone next year: brother, welcome to the future. my Instinct agent sends a message to your Muse agent, which alerts our friend's Grokbot, which gets picked up in our other friend's overnight Codex sweep and fed back to my agent as something I might find interesting. none of us have read a message or seen in each other in months. group gets a 20 page recap that drops fridays. none of us read it. we're all paying $200 a month. are you ready for the future, anon? (sarcasm obvi, love these products)

Baseten’s CEO on why every company will want to own its AI / @tuhinone on the rise of AI agents, the demand for inference, and building a new kind of hyperscaler. Lately, I’ve been spending a lot of time thinking about the AI inference market and how big it could get. Agents like Muse, Instinct, Town, and Grok Bot aggressively use browsers, run their own computers, and burn through far more tokens than a traditional chatbot. As more people put them to work (Muse is number two in the App Store), the demand for inference, or the computing needed to run these models, should grow enormously. Tuhin and I discuss the rise of agents using browsers and virtual machines to get things done, and @baseten's recent acquisition to help power that shift. We also talk about the data center backlash, why companies are embracing Chinese open models, his plans for Baseten’s new research lab, and why he thinks inference becomes the only market left after AGI. Timestamps: 00:00 What Is AI Inference? 06:05 Competing With the Cloud Giants 09:24 Why Companies Want to Own Their AI 15:30 Building Baseten Before the AI Boom 23:45 DeepSeek and the Race for Open AI Models 29:32 Baseten’s Growth and Expansion 33:05 AI Agents and the Blaxel Acquisition 38:32 Data Centers and the AI Backlash 43:25 What Happens to Inference After AGI? 45:12 When AI Agents Become Customers Thanks to the show's premier sponsors: @Atlassian, @meetgranola, and @mercury.

We need to be building apps for Siri ai/instinct/muse/whatever that cal ai guy is building t.co/WP45NX6sVE

Say someone fills out your demo form right now. What happens next? I’ve been working on getting that first response down to under 60 seconds. Grok handles the first call and re-qualification. Meta Muse helps with prospecting and follow-up drafts. Instinct Agent keeps the conversation going over text. GPT-6 Astra helps me spot where leads are falling through the funnel. I’m putting the full system together live this Wednesday so you can see how the pieces work together, what I’d automate, and where I’d keep a human involved. Wednesday at 1:30pm CT / 11:30am PT. Free. Register below