AI Agents Instinct and Meta's Muse Now Autonomously Make Phone Calls and Perform Real-World Tasks

The competitive landscape for autonomous AI agents has intensified with the simultaneous announcement that Instinct and Meta’s Muse are now capable of initiating and conducting phone calls on behalf of users, handling tasks from securing restaurant reservations to canceling subscriptions. This pivotal development elevates these agents beyond mere conversational interfaces, positioning them as proactive digital proxies designed to execute real-world actions.

Instinct, an invite-only personal AI agent from Spear Street Technology, Inc., distinguishes itself by operating without a dedicated app interface; users interact with it primarily via text messages or phone calls, delegating tasks directly. Launched in late August 2026, Instinct connects deeply to a user's digital life, including email, messaging, screen, audio, and location, leveraging a persistent cloud computer with browser access and cached credentials to perform complex, multi-step actions. Early adopters have reportedly used it for diverse tasks such as booking medical appointments, negotiating bills, monitoring ticket availability, coordinating vendors, purchasing groceries, and cancelling subscriptions. The company, led by 23-year-old Noah Shinn, rapidly achieved a $2.5 billion valuation after securing $250 million in Series B funding, bringing its total funding to $350 million. However, this deep integration has raised privacy concerns, with reports of retained email data, emails sent without explicit user approval, and successful prompt-injection tests, prompting Instinct to revise its terms and introduce a deletion tool in August 2026. Its terms of service grant a broad, perpetual license over user inputs and outputs for model training, with a notable exclusion only for Google Workspace data received through Google's APIs.

Concurrently, Meta’s Muse, launched in the US on September 8, 2026, across iOS, Android, and WhatsApp, presents itself as a personal AI agent built for everyone, focusing on taking actions rather than merely answering questions. Powered by the Muse Spark 1.3 model, Muse aims to tackle larger goals beyond one-off tasks, managing activities like sending emails, booking travel, filling out forms, and tracking long-term objectives. It operates within a dedicated, isolated cloud environment called the Muse Secure VM, where a "Sentinel" gatekeeper agent approves all outbound actions, such as sending emails or making purchases. Meta has implemented several privacy safeguards, including no direct visibility into user passwords or payment methods, the use of one-time virtual cards for purchases, and an option for users to opt out of having their interactions used for model training. While Meta states Muse is free for most everyday use, it offers paid subscription tiers—a Power plan at $20/month and a Maximum plan at $100/month—which primarily offer increased agent usage rather than additional features. Despite these measures, early testers reportedly encountered unexplained disconnections and instances where private photos were surfaced.

This advent of truly agentic AI, capable of autonomous outbound communication, marks a significant inflection point in the consumer technology landscape. For users, the promise is unparalleled convenience, offloading mundane yet time-consuming tasks that previously required direct human intervention. The "no new interface" approach of Instinct, in particular, lowers the barrier to adoption, integrating seamlessly into existing communication habits. This shift allows individuals to reclaim valuable time, transforming personal productivity and leisure by delegating tedious administrative burdens. However, it also necessitates a new level of trust and vigilance, as users empower AI with the authority to act on their behalf, often involving sensitive personal and financial data.

For the industry, this represents a profound transformation, particularly for customer service and contact centers. Gartner predicts that agentic AI will autonomously resolve 80% of common customer service issues by 2029, a dramatic increase from current figures. This translates to a projected $80 billion in contact-center labor savings in 2026 alone. The role of human agents is evolving from handling repetitive queries to focusing on complex, empathetic interactions that require judgment and creative problem-solving. This competitive push by Meta and Instinct accelerates the "AI-first with a fast human handoff" model, which is rapidly becoming the industry standard.

Compared to prior generations, such as Google Duplex, which pioneered human-like AI calls for restaurant bookings and salon appointments starting in 2018, these new agents demonstrate greater breadth of action and deeper integration into a user's digital ecosystem. While Duplex focused on specific "closed domains," Instinct and Muse aim for a more generalized agency, capable of navigating web forms, negotiating, and even managing long-term goals. Established rivals are also advancing: Amazon’s Alexa+, rolled out nationwide in February 2026, now functions as a sophisticated conversational partner for complex tasks and proactive assistance, powered by large language models like Nova. Apple's Siri AI, introduced in June 2026, is deeply integrated across Apple's ecosystem, leveraging Apple Intelligence for personal context understanding and actions across apps. The current wave of AI agents, however, takes a more comprehensive, persistent, and proactive approach to task execution across disparate services.

Looking ahead, the trajectory is clear: AI agents will continue to gain autonomy and sophistication. We can anticipate even greater multimodality, with agents integrating visual intelligence (as seen in Apple’s Siri AI) and other sensory inputs to better understand context. The tension between general-purpose agents and specialized AI solutions for specific industries will likely persist, with both evolving to meet distinct user needs while adhering to domain-specific compliance. The emphasis on robust security, verifiable actions, and granular user controls, exemplified by Muse's Sentinel gatekeeper and one-time virtual cards, will become paramount as these agents handle increasingly sensitive tasks. Furthermore, the legal and ethical frameworks surrounding AI autonomy, transparency (mandating disclosure when interacting with AI), and data governance will undergo rapid development, driven by the escalating capabilities and real-world impact of these agentic systems. The era of AI not just assisting, but actively *doing*, has undeniably arrived, and its implications will reshape our digital and personal lives for years to come.