On November 12, 2026, Meta opened outbound business calling to every paid Muse subscriber in the United States — reservations, cable holds, DMV-style chores — after a September beta that required users to ask Muse for access. The shift is less about a new capability than monetization and scale: phone tasks become the reason to pay $20 or $100 per month on top of a free tier that Wang says will soon get Muse email, but not calls.
For builders, the story is agent telephony as a paid SKU, sitting beside PayPal and Shopify transaction flows and the human-concierge experiment Meta rolled back.
TL;DR
| Question | Answer (Nov 12–13, 2026) |
|---|---|
| Who can use it? | Paid Muse subscribers in the US, per TechCrunch and Wang's X post |
| Price floor | $20/mo Power plan — 500M Muse tokens/week on Meta Help Center |
| High tier | $100/mo Maximum plan — 3B tokens/week (Help Center and TechCrunch agree) |
| Free tier | No calling; email address feature promised for free users later |
| Rivals | Instinct Concierge, Google Gemini business calling (TechCrunch) |
| Human operators? | Not addressed in Nov 12 launch copy; September Reuters test was rolled back |
| Receiving-end risk | Businesses may hang up on obvious AI callers (TechCrunch) |
What changed on November 12
Meta tested outbound calls in mid-September 2026 with users who explicitly requested access — Ryan Fox described prioritizing people who asked their Muse to opt in. November 12 removes that gate for any paying customer in the US.
TechCrunch's Sarah Perez quotes Alexandr Wang pitching insurance, cable, DMV, and restaurant reservations. Wang heads Meta's Superintelligence Labs and has been listing Muse launches publicly since connectors, Muse for Mac, and small-business tooling.
Meta also claims 10 million Muse users, a figure TechCrunch repeats from prior Meta commentary. That number mixes free and paid; calling is the nudge toward subscription revenue after reported DAU plateaus in early October analyst notes.
Pricing and what Meta documents
Official pricing lives on Meta's site, not in the TechCrunch article alone.
Meta's Muse subscriptions page lists:
- Power plan — $20/month — 500 million Muse tokens per week
- Maximum plan — $100/month — 3 billion Muse tokens per week
Subscriptions renew monthly, require a Meta account, and vary by region during limited rollout. Neither the Help Center nor the subscription onboarding text we could read names "phone calls" as a bullet feature — only token buckets. That matters for finance teams: telephony cost is embedded in token usage, and Meta has not published per-call metering.
Meta also sells additional Muse tokens as non-refundable, non-transferable entitlements (token purchase help). Heavy callers on the $20 plan may hit weekly caps before they finish a long hold.
Disclosure, humans, and trust
Three threads collide:
- September human concierge — Reuters reported Meta routed some calls through trained contractors with high success rates but weak disclosure. Meta rolled it back after internal privacy objections.
- November launch silence — TechCrunch does not say whether contractors returned. If you ship a competitor agent, assume users will ask the same question.
- Callee identification — Engadget's prior coverage noted demos where Muse said it was calling on someone's behalf and transcribing without always saying "I am an AI." Synthetic-voice policy posts already flag outbound-agent abuse; Muse pushes the same regulatory edge for ordinary customer service.
Meta's commercial terms still make customers solely responsible for AI mistakes. A misheard reservation or leaked account number on a call is user liability under published terms — awkward next to enterprise Muse sales motion.
On-device phone assistant orb symbolizing Muse placing outbound business calls
Competitive landscape: Instinct, Gemini, Grok
Instinct shipped Concierge calling earlier in the fall; TechCrunch framed September as feature parity among consumer agents. Google Gemini offers a similar business-call option per the same reporting.
Our Dots vs Grok vs Muse comparison already treated voice and computer-use as the differentiation axis. November 12 makes telephony table stakes, not a novelty.
What still differs:
| Dimension | Muse | Notes |
|---|---|---|
| Paywall | Calls require $20+ sub | Grok and others gate different perks |
| Commerce graph | PayPal, Shopify, Expedia hooks | Instinct leans vault + 1Password narrative |
| Enterprise path | IBM / Accenture pipeline reporting | Gemini rides Workspace distribution |
| Trust week history | Human concierge + Mac token hijack | See security week post |
Chat window with ad-funded tokens illustrating paid Muse subscription gating for phone calls
What people are asking
Will my restaurant stay on the line? TechCrunch is blunt: many receptionists hang up on AI. Success may depend on voice realism and how fast the callee detects synthesis — the same tension contact-center voice models solve with latency budgets Muse does not publish.
Does this work in Canada or the EU? Wang's announcement specified US paid subs. Meta Help Center still says Muse and subscriptions are in limited geographic testing. Canadian users on Techmeme asked for parity; no official expansion date appeared in sources we read.
Is this a regulatory target? US state two-party consent recording laws, FTC expectations on AI disclosure, and PCI rules if callers read card numbers aloud all apply to your workflow if you connect billing systems. Muse's September human-operator debate showed Meta knows disclosure is radioactive.
Should my startup build calling too? If you compete on vertical tasks (clinics, trades, insurance FNOL), you may beat general Muse on scripts and CRM writes. If you compete on consumer lifestyle, Muse's distribution and Meta account graph are hard to match. Personal agent protocols matter if businesses want one integration surface.
Wang's announcement and what Meta did not say
Alexandr Wang's November 12 post on X shows the consumer framing Meta wants: Muse as relief from calls users dread. TechCrunch embedded the message; explainx.ai could not retrieve a working oembed response for the status URL during validation, so rely on TechCrunch's republication for verbatim wording.
What is missing from the launch communications we read:
- Per-call limits — no cap on concurrent calls, daily dial count, or maximum hold time.
- Recording and retention — transcription is mentioned in older demos, but November 12 copy does not say where audio is stored or for how long.
- Business allowlists — no statement on whether Muse can dial toll-free only, mobile numbers, or international extensions.
- Success metrics — Meta does not cite completion rate for the broad paid rollout, only anecdotes and prior beta claims.
That gap matters because OpenAI's workplace agent push and Amazon Bedrock agent runtimes are racing on measurable task completion, not demos. Phone calls are the first Muse feature where third-party humans grade the agent in real time — receptionists, not benchmarks.
How to try it (US subscribers)
Meta documents subscription purchase on mobile and at muse.ai:
- Open Muse, go to Settings → Upgrade, pick Power or Maximum (subscribe help).
- Confirm you are in a supported US region during limited rollout.
- Ask Muse in natural language to place a business call (restaurant, insurer, cable ISP).
- Review what data Muse can already read via connectors — calls inherit that context.
Meta does not publish API access for third parties to trigger Muse calls programmatically in the sources we checked; this remains a consumer product surface, not a telephony API like Retell's Dialer stack.
Builder takeaways
- Paid-only telephony tests whether users value time saved on hold more than chat tokens — a pricing experiment other agent apps will copy.
- Token-priced calls hide telephony COGS; watch for throttle complaints if weekly 500M tokens cap bites mid-task.
- Callee hang-ups are a product metric, not a footnote. Benchmark your agent against human success rates Meta once claimed for contractors (95–98% in Reuters reporting) versus AI-only floors.
- Enterprise buyers should demand written answers on human involvement, recording, and data retention before employees connect work inboxes — align with Muse enterprise liability language.
Why this launch timing matters for Meta's agent bet
Meta Connect positioned Muse as the consumer face of Superintelligence Labs messaging. Voice calling is the first feature that outsources social friction — negotiating with a human stranger — rather than summarizing email or editing photos.
That is strategically closer to Instinct's concierge identity than to Meta AI inside Instagram feeds. It also intersects with ads: Wang has said Muse stays ad-free for now, but paid tokens recycle the ad-funded playbook into subscription ARPU. If calling converts skeptics who plateaued on chat, Meta can argue agents deserve Super Bowl-level spend without reopening the daily user plateau debate.
For regulators, the launch lands beside Google Classroom AI protests and broader agent permission fights. Muse calls do not need school board approval, but they do need consumer understanding of who listens. Meta's September rollback of undisclosed humans is the precedent; November silence is the open question.
Related reading
- Meta Muse human concierge and Mac zero-day trust week
- Muse transaction layer: PayPal, Shopify, Expedia
- Meta Muse DAU plateau: how to read the numbers
- Dots vs Grok vs Muse agent comparison
- Instinct and 1Password vault narrative
- Synthetic voices, swatting, and telephony policy
- Meta enterprise sales and Muse business agents
Muse plans, regions, and features are accurate as of November 13, 2026, per Meta Help Center and TechCrunch.
