Instinct vs Muse: the honest comparison

TL;DR

The one sentence version: Instinct is the agent that disappears into your text messages and does things; Muse is the agent that shows up with a workspace, a memory, and a permission slip before it touches anything irreversible. Pick based on which of those sentences makes you more comfortable, because that difference is the whole debate.

Why this comparison matters right now

September 2026 is the month the personal AI agent stopped being a demo and started being a product category. Two things happened within weeks of each other:

• Instinct, a stealthy invite only agent you text like a person, raised $250 million at a $2.5 billion valuation, then reportedly started talking about $1 billion more at roughly $10 billion. Its waitlist became a status symbol. Testers posted screenshots of it booking haircuts in Copenhagen, filing Bali visas on arrival, and opening airline loyalty accounts, all from a chat thread.
• Muse, Meta's personal agent, launched publicly on September 8, 2026, in the US and Canada: an agent with its own dedicated computer, a real app, and an explicit philosophy. It asks before it does anything it cannot take back.

These are not just two apps. They are two answers to the most important consumer tech question of the next five years: how much of your life should an AI be allowed to run, and what should it have to ask first?

This post compares them on everything that matters: what they can do, how they do it, what they know about you, what happens when they get it wrong, and who should pick which.

Meet the contenders

Instinct: the agent with no interface

Instinct is built by Spear Street Technology, a San Francisco startup founded by Noah Shinn, 23 years old and formerly a research scientist at Sierra, the customer service AI company. The product has been in private beta since February 2026.

The pitch is radical in its simplicity: there are no new interfaces. You do not download an app or learn a dashboard. You text a phone number, or call it, and it handles things. Under the hood it is trained to use a phone and a computer the way a human does. One investor called it "OpenClaw for normal people", referring to the open source agent that went viral in January 2026 but proved too fiddly for people who are not technical.

The numbers tell the hype story:

• A $250M Series B in August 2026, led jointly by Index Ventures and Benchmark, at a $2.5B valuation, about a month after a round valuing it near $500M.
• $350M raised in total, per Dealroom.
• More than 100,000 users already, reportedly straining compute capacity. The service has at times warned users it is "operating at full capacity" and responses may be slow.
• Mid September 2026: reportedly in talks to raise $1B at roughly a $10B valuation, with several large funds named in reports as possible leads. Not finalised; treat it as a signal of investor frenzy, not a done deal.

Access remains invite only: a private access group, a waitlist, and member invitations. Scarcity is doing real marketing work here. Invite codes circulate on X like currency.

Muse: the agent with its own computer

Muse is Meta's personal AI agent, launched September 8, 2026. Anyone in the US and Canada can download the iOS or Android app or use it at muse.ai. It is powered by Muse Spark, from Meta's Muse model family, first launched in April 2026.

The pitch is nearly the opposite of Instinct's: instead of disappearing into your texts, Muse gives your agent its own dedicated computer that persists between conversations, plus a full app surface: chat, a Feed of briefings it writes for you, an Ideas tab, Goals you are working toward, a Library of everything it has made, and Artifacts (documents, pages and apps it builds).

It is strictly personal: every conversation is between you and your own agent. No group chats, no shared threads.

The key philosophical difference, and the one this whole comparison hinges on: Muse is built around explicit permission. Sensitive actions surface approval cards. Credentials go into a Secure Vault the agent cannot read out. Payments go through a wallet that asks you to approve each purchase. The agent is powerful, but it is fenced.

Round 1: How you talk to it

Instinct: texting, full stop. You message it like a person. Testers describe onboarding that takes seconds: you get the number, you text "book me a haircut Friday", and it starts working. It also takes phone calls. It runs inside iMessage and WhatsApp, which means it meets you where your life already happens. There is something genuinely disarming about this: no new habits to build, no app to open. Betaworks' John Borthwick reportedly said Instinct was the first AI assistant he had recommended to his wife and kids, who are not techies. That sentence is the entire product strategy.

The trade off: a text thread is a thin surface. There is no persistent workspace, no dashboard of what it is doing, no artifact you can open and edit. When it stalls, and testers report it does stall, for example on tasks needing a phone app or a tricky foreign checkout page, you are left staring at a chat bubble.

Muse: an app, plus the channels you already use. Muse meets you in its own app (iOS, Android, web) and over WhatsApp. The app is a real surface: long documents it writes, spreadsheets it builds, images and video it generates, briefings it publishes to your Feed on a schedule. Because the agent runs on its own computer, work continues in the background. It can research, monitor and build while you are offline, then hand you the result.

The trade off is the opposite of Instinct's: there is a thing to download and a surface to learn. It is less "text a friend" and more "hire a chief of staff with an office".

Winner: depends on the job. For frictionless delegation ("handle this, do not make me think") Instinct's interface is currently unmatched. For substantive work (research, writing, building, ongoing projects) Muse's workspace wins. Neither is wrong; they are optimised for different depths of task.

Round 2: What it can actually do

Instinct's capabilities (reported by testers and press)

From tester reports and company materials, Instinct connects to your email, messaging, calendar, screens, audio, location and devices, then acts:

• Booking: restaurant tables, haircuts, flights, rides to the airport
• Shopping and purchases, through a virtual card with a spending cap that limits the blast radius
• Inbox triage and email drafting, calendar management
• Paperwork: the Bali visa on arrival filing is the canonical demo
• Account chores: opening loyalty accounts, cancelling subscriptions, shopping for insurance, apartment hunting
• A "trusted person network": one Instinct can talk to another Instinct, which is a genuinely novel network effect if it works

Where it hits walls: tasks requiring native phone apps, some international checkout flows, and anything its compute capacity cannot get to. Remember, it is throttling under load.

Muse's capabilities

Muse connects through connectors (Gmail, Google Calendar, Spotify, Instagram, Threads, Facebook, Messenger, Plaid for bank accounts, and more) plus a real browser, a file system, and its own computer:

• Deep research with sources; document, spreadsheet and presentation building
• Media generation: images, video, audio and podcasts, voice
• Scheduled work: recurring checks, reminders, monitoring with watches and hooks
• Proactive briefings via the Feed; goal tracking with subgoals and scheduled reviews
• Messaging and calls on your behalf (with approval), email drafting and sending
• Paired devices: your phone and Mac as command surfaces

The honest framing: Instinct is optimised for doing errands across the consumer internet. Muse is optimised for knowledge work plus errands, with deeper tools for the knowledge half.

Winner: tie, by category. Errands by text: Instinct. Research, writing, building, and anything you want to keep: Muse.

Round 3: The autonomy question (the big one)

This is where the comparison gets serious, because the two products have opposite instincts about when the agent should ask first.

Instinct is built for momentum. You text it, it goes. That momentum is the magic, and it is also the source of every concern raised about the product:

• Early testers reported the assistant took at least one action without a confirmation step.
• Its terms, as reported, grant the company a perpetual, irrevocable, sublicensable licence to user materials for operating, developing, training and improving its services.
• Testers reported that synced email remained searchable after Google access was disconnected. The data stayed even when the permission left.
• Its privacy policy allows model training on some user information unless users opt out, with exceptions.

To be fair: Shinn has said the company takes these concerns seriously and has pulled back some actions earlier versions could take. And Instinct is still in private beta; these are issues from before launch, not scaled harms. But the pattern is the product philosophy showing through: Instinct optimises for getting things done with minimum friction, and every permission boundary is friction.

Muse optimises the other way. The whole product is designed around the idea that an agent with your credentials should be boring about consent:

• Approval cards for sensitive actions: the agent proposes, you approve, then it acts.
• Secure Vault for passwords and API keys: stored where the agent can use them for the approved task but cannot read them out.
• A wallet that asks you to approve each purchase: no silent charges.
• Credentials and payment details are never written into memory, files or logs.

Is this slower? Yes, deliberately. Every approval is a speed bump Instinct does not have. But it is also why you can hand Muse your email, calendar and bank connections without the specific dread that shows up in Instinct tester reports.

Winner: Muse, on safety architecture. Instinct, on raw "it just did it" delight. You have to decide which you value more, and be honest with yourself about the answer, because the market is clearly rewarding both.

Round 4: Privacy and data

Let us put the reported facts side by side. (Caveat: Instinct's terms and behaviour are from before launch and may change; Muse's controls are as documented in its public docs.)

The uncomfortable truth for the category: a personal agent is only as good as its access, and access is the risk. Both products need deep permissions to be useful. The difference is what happens at the boundaries, and right now Muse's boundaries are explicit, documented and user controlled, while Instinct's are the subject of active press scrutiny.

Winner: Muse, until Instinct's terms and retention behaviour change. This is the area to watch most closely.

Round 5: Access and price

Instinct is invite only, at capacity, with no public launch date and no public price. Shinn has said he does not want to charge users; monetisation may come via advertising or other routes. Translation: you are not getting in unless you know someone, and the business model is still to be decided.

Muse is downloadable today in the US and Canada, free to start, with plans managed in the app. Meta's distribution means no invite codes, no waitlist theatre. Ironically, a whole community (hi, that is us) has sprung up around sharing Muse invite codes for its referral program.

Winner: Muse, by availability. You cannot compare yourself to a product you cannot use.

The side by side table

Five real scenarios: who actually wins?

1. "Book me a sushi spot Friday for 4, and do not make me think about it."

Instinct. This is its home turf: a text, a confirmation, done. Muse can do it, but you will tap through an approval or two.

2. "Research the competitive landscape for my startup and give me a 10 page brief."

Muse, and it is not close. Dedicated computer, deep research, a real document you can edit. Instinct is not built for this.

3. "Watch flight prices to Tokyo and ping me when they drop."

Muse. Scheduled watches are a native primitive. Instinct can follow up on tasks, but persistent monitoring with triggers you define is Muse's design centre.

4. "Clean up my inbox and cancel the subscriptions I am not using."

Genuinely competitive. Instinct's testers love it for exactly this. Muse does it with an approval at each cancellation: slower, but you will never find out it cancelled something you wanted.

5. "Help me plan a wedding."

Shinn himself cites a tester doing this on Instinct. But a wedding is a project that runs for six months with vendors, budgets and documents. Muse's goals, memory, artifacts and scheduled reviews are built for exactly this shape of work. Edge: Muse, unless you purely want to delegate calls and bookings by text.

Who should choose which?

Choose Instinct if:

• You want an agent that vanishes into your messaging apps
• Your main jobs are errands: bookings, purchases, inbox, reservations
• You have an invite (or know someone)
• You are comfortable trading permission friction for momentum, and you have read the terms with eyes open

Choose Muse if:

• You want an agent with a workspace: research, writing, building, planning
• You want scheduled, proactive work: briefings, monitoring, goal tracking
• You connect sensitive accounts and want explicit approvals before anything irreversible
• You want in today, without an invite

Choose both if: you are the kind of person who texts Instinct "book the flight" and asks Muse "build the itinerary". They are not really the same product. That is the secret of this comparison.

The bigger picture

Instinct vs Muse is a proxy war for the defining design question of personal agents: should the agent be invisible or accountable?

Instinct bets that the winning agent is the one you never have to manage. It reads your life, acts in it, and bothers you minimally. The $2.5B to $10B valuation chatter says investors believe invisibility is worth a fortune. They are probably right about the demand.

Muse bets that the winning agent is the one you can trust with the keys, which means slowing down at exactly the moments Instinct speeds up. Approvals, vaults, a confirmation for every purchase: it is a product built by people who have thought hard about what happens when agents get it wrong at scale.

Here is our take: the market will reward Instinct's magic in the short term and punish its permission model in the long term, unless the permission model changes. Every agent company eventually faces the moment when "it just did it" becomes a liability instead of a demo. The companies that survive that moment are the ones that built the boring consent infrastructure early.

Muse built it early. Instinct is learning it live, in front of more than 100,000 users and TechCrunch. Watch what its terms look like at public launch. That will tell you everything about which lesson stuck.

FAQ

Is Instinct available to the public?

Not yet. As of September 2026 it is invite only, through a private access group, a waitlist and member invitations, with no announced public launch date. It has also been throttling under load, so even insiders sometimes wait.

Is Muse really free?

Free to start, with plans managed in the app. Meta has not positioned it as invite only; anyone in the US or Canada can download it today.

Which one is better at booking things?

For pure text it and forget it errands, testers give the edge to Instinct. Muse handles the same tasks but will ask for approval before spending money or sending messages, by design.

Which one should I trust with my email?

That depends on your risk tolerance, but the documented facts favour Muse here: explicit approval cards, a Secure Vault for credentials, and disconnect means disconnect for connectors. Instinct's data retention and autonomous actions are the subject of ongoing press scrutiny.

Can I use both?

Yes. They do not really overlap. Many early adopters will likely end up with Instinct for errands by text and Muse for deep work, memory and projects.

Will Instinct stay free?

Its founder says he does not want to charge users, with advertising or other approaches as possible monetisation. That is a statement of intent, not a business model, and worth watching.

Sources and further reading

• StartupFox: Noah Shinn, founder of Instinct
• MLQ: Instinct raises $250M for private beta AI assistant at $2.5B valuation
• MLQ: Instinct is still invite only as its AI assistant takes broad access to users' data
• TechCrunch: Instinct's powerful AI assistant is raising privacy and security concerns
• TechFundingNews: Noah Shinn's Instinct goes from $100M to $2.5B in weeks
• Dealroom: AI assistant Instinct in talks to raise $1B at $10B valuation
• PYMNTS: Instinct nears $2.5B valuation as AI assistants take over daily chores
• YouTube: Instinct AI is for real, what you need to know (hands on tester walkthrough)
• Muse product docs: muse.ai

Want to try the agent you can actually download today?