A new wave of artificial intelligence assistants can now make phone calls to businesses on a user’s behalf, handling tasks like booking restaurant reservations, joining waitlists or disputing a cable bill without the user ever picking up the phone.
Instinct, an AI assistant company, announced the feature, called Instinct Concierge, on Sept. 16. Founder Noah Shinn said the tool could call a restaurant that does not take online reservations, get a user onto a dentist’s cancellation list, or try to resolve a cable billing dispute. The feature is currently in early access, with plans to expand over time.
Meta has been testing a similar outbound-calling capability through its own AI agent, Muse, launched Sept. 8. The calling feature lets some Muse users ask the assistant to contact U.S. businesses directly, though it remains limited to testing rather than widely available. Muse itself is now available in the U.S. and Canada and can browse the web, fill out forms and keep working after a user closes the app.
Both companies are backed by serious money. Instinct raised $250 million in an August Series B round valuing the company at $2.5 billion, then announced a $1 billion Series C round on Sept. 28 from investors including Sequoia Capital, Benchmark and Coatue, pushing its valuation to $10 billion. Muse, meanwhile, climbed to the top of Apple’s U.S. App Store after launch, with reported downloads topping 3 million by late September.
What It Costs
Instinct is currently available only through private access, and the company has not published a monthly subscription price or separate pricing for the Concierge calling feature. Muse offers a free tier with usage limits; paid plans have been reported at $20 a month and $100 a month depending on usage, though Meta has not announced separate pricing for its phone-calling test feature.
The Access These Assistants Need
To make a call on someone’s behalf, these assistants typically need access to connected apps, messages, emails, voice data, account credentials and sometimes payment or health information. Instinct’s privacy policy confirms its assistant can access such information when a user grants permission, and that health-related data could enter the system if, for example, a user asks it to book a medical appointment.
Instinct says users can opt out of having certain information used to train its AI models, though that choice applies only going forward — models already trained using a user’s data may retain what they learned. Information flagged for safety review can still be used for AI training related to harmful content detection. Disconnecting a third-party integration does not automatically delete previously collected data; users must use a separate deletion control to erase it. Data received directly through Google Workspace APIs, however, is not used for training.
Meta says Muse runs inside a dedicated secure virtual machine in the cloud and cannot see passwords or payment methods stored in its secure credential storage. The assistant asks for approval before sensitive actions like sending an email or making a purchase, and keeps an audit trail of what it has done and plans to do. For purchases, Meta says its Link feature can generate a one-time card number so the user’s actual card is not exposed to the merchant or the AI. Muse conversations and data are not shared with Meta’s advertising systems, and users can opt out of having interactions used for model training.
Where the Risk Lies
Unlike a chatbot answer a user can read and correct before acting on it, these AI agents take action directly — and errors may not always be reversible. Instinct’s own terms acknowledge its services can produce incorrect or incomplete information and that actions taken on a user’s behalf
You must be logged in to post a comment Login