Meta Connect 2026 (September 23–24, Menlo Park) was Mark Zuckerberg's product keynote for Muse, Ray-Ban Meta glasses, Muse Charm, and Meta VR Glasses. This recap lists every launch, what personal AI can do after Connect, where Meta goes next, and what developers can build on Muse connectors and the Wearables Device Access Toolkit.
$349 Ray-Ban Meta Audio, October 13 · $449 Gen 3 · $249 Adventurer · $1,299.99 VR Glasses, Spring 2027.
If you build, the question is no longer "can I wrap an LLM." It is: what verb does Muse still lack?
Updated September 24, 2026.
What actually happened to Muse
Muse launched weeks before Connect and already behaved like a personal agent: goals, apps, background work. Connect added the missing pieces that turn a chat product into infrastructure.
It can talk and work at the same time. Realtime voice is not a dictation feature. You keep talking. Muse keeps executing. That is the difference between an assistant and an employee.
It got a face. Muse Realtime Avatar turns the agent into a character you can video-call. Default is Jolly. Custom voices, custom look. Soft, but it is how consumers will bond with an agent they let shop and email.
It got an inbox. Muse has its own email address. Other people and services can message the agent, not you. That is how an agent becomes a coworker instead of a search box.
It got a computer. On Mac, with permission, Muse drives apps after you walk away. Files, mail, calendar, browser. The agent is no longer trapped in its own UI.
It got eyes. Muse is coming to Meta AI glasses. Wake it by name. It can act on what you are looking at — a product on a shelf, a flyer, a school-supply list — without you describing the scene. That is the whole interface shift.
It got a pocket. Muse Charm is a keychain holdable: small screen, mics, fingerprint wake, full voice + avatar. Not a phone. Not glasses. Always-on talk surface.
It got a business model. Free for now. Later Meta takes a small fee on transactions. Muse is being built as a commerce and work runtime, not a $20/month chatbot.
It got a platform. Connectors: Walmart, Best Buy, Gap, Sephora, Ulta, Wayfair, Shop Pay, PayPal, Shopify catalog, Expedia and Instacart soon, Notion, GitHub, Box, Spotify. Meta opened developer submissions. Thousands of apps in days.
That is what happened. Muse stopped being a destination app and became a layer that other products plug into.
The stack Connect made real
MUSE SPARK (see, reason, act)
↓
MEMORY (you, goals, history)
↓
CONTEXT (glasses camera/audio, Charm, Mac screen)
↓
AGENT (Muse)
↓
ACTION (connectors, payments, computer use)The opportunity is not "train a bigger model." The opportunity is every layer Meta will not finish itself: niche actions, trust, vertical memory, and glasses-native workflows.
What personal superintelligence opened (this is the market)
1. Agent verbs — the biggest new market
Muse can only do what a connector lets it do. If your product is a website, you are invisible. If you are a verb — reorder, file_claim, book_tech, pull_labs, close_ticket — you exist inside someone else's day.
Build: Muse connectors for things Meta will never first-party.
- Local services: plumbers, clinics, tutors, landlords
- SMB ops: invoices, inventory, scheduling, payroll
- Vertical data: legal dockets, insurance, property, labs, auto parts
- Post-purchase: warranty, returns, reorder, "fix this bill"
Submit early. Meta is onboarding in waves. First useful connectors become the default skill for that category.
2. Agent commerce
Muse can search Shopify and check out with Shop Pay / PayPal. Retail connectors are landing. The session no longer starts on your homepage. It starts with "Muse, get me the one that fits."
Build:
- Catalogs agents can query (size, stock, substitutes, constraints)
- Return / exchange APIs
- "Buy the replacement" flows (broken vase demo is the template)
- Merchant-side guardrails: spend caps, approval on big carts
Whoever owns agent checkout owns the cart that never opens a browser tab.
3. First-person software (glasses)
Old apps wait for a typed prompt. New apps run on what the user is looking at.
Muse on glasses + Device Access Toolkit + Display Web Apps = a new UI class.
Build:
- Look-and-act tools: identify, price, log, inspect, annotate
- Field work: warehouses, job sites, clinics, classrooms
- Training overlays: form check, recipe steps, repair guides
- Accessibility beyond Meta's hearing mode
- Voice-only utilities for camera-free Audio glasses ($349, all-day wear)
If your app still needs the user to narrate the scene, you are one generation behind.
4. Background work while the human leaves
Mac computer use + email address + connectors means Muse can run a loop you are not watching.
Build:
- "Muse, handle this thread" products (refunds, scheduling, vendor chase)
- Audit trails: what it clicked, what it spent, what it sent
- Human-in-the-loop approvals for money, legal, health
- Small-business operators: inbox, books, restock, follow-ups
The product is not chat. The product is a finished task when they look back.
5. Personal memory layers
Meta will store some memory. Users will not want all of it in one company.
Build:
- Private memory vaults Muse can query with permission
- Team / family shared context ("our house, our kids' school list")
- Work memory that never leaves a VPC
- Preference graphs: fit, diet, budget, brands to avoid
Personal superintelligence is only as good as the memory it is allowed to use. That layer is underserved.
6. Trust is an unbuilt product category
Glasses plus an agent that can spend money will scare people. Meta's answers (Audio glasses, Secure VM, some on-device processing) are not a full product.
Build:
- Permission dashboards: see / buy / send / drive-my-Mac
- Per-connector spend limits
- Workplace camera-off policies
- Enterprise connectors with logs
- "Explain why Muse did that" receipts
Trust software will sell to the same users who download Muse and then get nervous.
7. Spark / Muse Code as internal infrastructure
Spark is the multimodal acting model. Muse Code is the repo agent. Sessions at Connect treat Spark as perceive → decide → act, not caption.
Build:
- Internal agents with harnesses: plan, patch, test, roll back
- Video-heavy workflows (support, QA, field clips from glasses)
- Voice-transcribe → ticket → fix loops
Do not sell "a coding chatbot." Sell closed loops with an audit.
What not to build
- Another general assistant
- A Muse skin with no unique verb
- A glasses app that is just a phone UI shrunk to 600px
- Shopping search with no checkout
- Memory features that silently ship user life to a third party
The platform already has a general agent. It needs specialists it can call.
Builder order for the next 30 days
- Pick one verb Muse cannot do well in your industry.
- Ship a connector. Make the action reliable, not clever.
- Add a glasses or voice path only if the verb needs eyes or hands-free.
- Wrap money-moving actions in an approval + log.
- Publish "works with Muse" as the product page, not "powered by AI."
The opening, in one line
Connect did not finish personal superintelligence. It opened the API around it.
Muse is the agent. Spark is the brain. Glasses are the context. Charm is the always-on mic. Connectors are the market.
The companies that win are the ones that become a sentence Muse can say out loud:
"I already handled it."
Meta Connect 2026 FAQ
What did Meta announce at Connect 2026?
What is Muse at Meta Connect 2026?
How much are Meta VR Glasses?
When do Ray-Ban Meta Audio glasses ship?
Source: Meta's official Connect 2026 recap.
More from techwarq: Jev is a smart if-statement.