Meta’s Vishal Shah on Muse: How ‘Personal Superintelligence’ Could Change Everyday AI

Meta is positioning its latest artificial intelligence assistant, Muse, as something more ambitious than a conventional chatbot.

The company wants Muse to understand a user’s context, goals and preferences and eventually perform tasks on the user’s behalf. The vision also extends beyond smartphones, with Meta planning to make its AI a central part of its smartglasses ecosystem.

Speaking to Financial Express at Meta Connect 2026, Vishal Shah, Meta’s Vice President of AI Products, discussed the company’s approach to personal AI, user control, autonomous agents and the development of Muse Spark 1.3.

Shah described the product as an early step toward what Meta calls “personal superintelligence.”

Muse Is Designed to Do More Than Answer Questions

According to Shah, Meta wants Muse to move beyond the traditional model of users typing questions into an AI chatbot and receiving responses.

The company is developing Muse to understand what users are trying to accomplish and help complete tasks that may otherwise require multiple steps.

New capabilities announced around Meta Connect include an updated Mac application that can interact with and operate a computer, along with voice functionality and an AI character powered by a new embodiment model.

Meta also plans to make Muse the primary AI agent on its smartglasses.

That would allow users to interact with their personal AI while away from a phone or computer.

Meta Wants Users to Stay in Control

Greater AI autonomy creates an important product challenge: deciding what an AI agent can do without permission.

Shah said Meta’s approach is based on keeping the user in control.

Certain actions, such as making a purchase or sending something on the user’s behalf, will require approval, according to the executive.

For other tasks, users can determine how much autonomy Muse receives.

Shah said the system initially takes a more conservative approach and can become more flexible as users become comfortable with it.

Users Can Set Permissions for Connected Apps

Meta is also building controls around the services that Muse can access.

Shah used email as an example. When Gmail is connected, users can choose whether Muse can only read messages, read and write messages, or request permission before individual actions.

The executive said these controls extend beyond the visible application interface.

According to Shah, Meta has also incorporated permission and security controls into the secure virtual machine used to operate Muse’s computer functions in the cloud.

The idea is to make security and privacy part of the underlying architecture rather than treating them as an additional layer added later.

Why Meta Starts With Conservative Defaults

Shah said an AI assistant can become frustrating in two opposite ways.

If it takes unexpected actions, users may feel that they have lost control. If it asks for permission too frequently, the experience can become cumbersome.

Meta’s approach is therefore to begin with more conservative defaults.

As users become comfortable with Muse, they can choose to give the system more autonomy.

Shah said the same principle applies to connectors and other services, with permissions expanding only when users choose to do so.

What Is Muse Spark 1.3?

Muse Spark 1.3 is the model Meta says it developed specifically for the types of tasks associated with personal AI and long-running agents.

Rather than optimising solely for benchmark-style intelligence, Shah said the model was trained around practical product use cases.

One example is a complex project that needs to be divided into multiple tasks. Muse could potentially break the goal into smaller pieces and use multiple subagents to work through them.

The model can also be used for coding.

Shah said coding capability is important because Muse may encounter software or services without a ready-made connector. In such cases, the AI could potentially create a connector itself by writing the required code.

Meta’s Vision: Personal AI for Everyone

Meta is also emphasising accessibility as part of its personal AI strategy.

Shah said the company believes personal superintelligence should be available broadly and pointed to Muse’s large free tier as evidence of that approach.

Meta could eventually introduce subscriptions or other commercial models for heavier usage, but Shah said the basic product should remain approachable.

The strategy reflects Meta’s broader consumer-platform model: build AI into products already used by large numbers of people rather than requiring everyone to adopt a separate specialised AI service.

From Phones to Smartglasses

Meta’s longer-term vision extends beyond screens.

Shah said smartglasses could connect the digital capabilities of an AI assistant with a user’s physical surroundings.

Instead of interacting with Muse only through a phone or computer, users could potentially access their AI while moving through the real world.

This could allow AI to assist with tasks that combine digital information and physical context.

Shah described the objective as helping people spend more time on things they want to do rather than tasks they simply need to complete.

These capabilities remain part of Meta’s evolving product roadmap rather than a fully realised future product experience.

Early Uses Could Include Finance, Health and Productivity

Shah said early Muse use cases include personal productivity, personal finance, and health and fitness.

He suggested that these categories could eventually develop into deeper AI-assisted experiences.

The broader ambition is for Muse to understand longer-term goals rather than simply respond to isolated prompts.

That would require the system to retain relevant context, coordinate multiple tasks and determine when user intervention is necessary.

What About Muse in India?

India is an important potential market for Meta’s AI ambitions, but Shah acknowledged that Muse needs further development before it can become broadly useful across the country.

At the time of the interview, Muse was limited to English.

Language support is therefore one of the areas Meta needs to expand. Shah also highlighted the importance of building connectors for the products and services commonly used in different markets.

Scaling the technology internationally is another challenge.

Meta’s stated objective is to take its products global, but Shah noted that rolling out a new AI product across different markets takes time.

Why Shah Sees Opportunity in India

Shah said he has a personal connection to India through his family and described the country as a significant entrepreneurial market.

He also pointed to India’s large population of small businesses and technology users as a potential opportunity for AI tools.

In his view, personal AI could help people overcome gaps in access to information, technology and other resources.

For Muse to reach that potential, however, Meta will need to address language support, local product integrations, availability and scaling.

Meta’s Personal AI Ambition Is Still at an Early Stage

Despite Meta’s long-term ambitions, Shah emphasised that Muse remains at an early stage.

The company is adding capabilities around computer control, voice interaction, AI embodiment and smartglasses while continuing to develop its underlying models and infrastructure.

The central idea is changing the role of AI from a tool that waits for instructions into an agent that understands context and can take action—with the user deciding how much autonomy it receives.

Meta’s success with that model will depend not only on the intelligence of Muse, but also on the reliability of its actions, privacy protections, permission controls and the usefulness of its integrations.

For now, Meta’s vision is clear: build a personal AI that can understand what users want to accomplish, help execute the work and eventually accompany them beyond the smartphone.