Category: Technology & AI

By Simon Ben Gemmill | 24 September 2026

The artificial intelligence race is entering a new phase. Instead of simply asking an AI chatbot a question and receiving an answer, technology companies are increasingly developing AI agents designed to carry out tasks on a user’s behalf.

Two of the biggest companies pursuing this idea are Google and Meta. Both are developing AI systems that can work across applications, understand context and perform multi-step tasks. The result could change the way people use computers, smartphones and the internet.

What is an AI agent?

A conventional chatbot generally waits for a question and produces an answer. An AI agent is designed to go further. An agent can potentially plan a task, use software, browse websites, interact with services and complete several steps before reporting back to the user.

For example, instead of asking an AI where to find a hotel in London, an agent could potentially search for hotels, compare options, check availability, prepare a booking and then ask the user for approval before completing the transaction.

This moves AI from being primarily an information tool towards becoming an action-taking digital assistant.

Meta’s Muse

Meta launched Muse on 8 September 2026 as a personal AI agent. Meta says Muse is designed to do more than answer questions and can work on tasks and projects on a user’s behalf.

Meta says Muse can send emails, book travel, fill in web forms and perform other multi-step tasks. It can continue working on longer tasks after a user closes the application and return when it needs approval for an important action.

Meta also says Muse uses a dedicated secure virtual machine, effectively a separate computer in the cloud, to isolate an agent’s activity. Users choose which applications Muse can access and can revoke access.

Read Meta’s official Muse announcement.

Google and Gemini

Google is approaching the same broad idea through its Gemini ecosystem. Google has been adding agentic capabilities to Workspace so Gemini can work across applications such as Gmail, Drive, Docs, Slides and Chat.

Google describes Gemini as an intelligent orchestrator that can help carry out complex tasks across several applications rather than requiring users to move manually between different programs.

Google is also developing more personal AI assistance and tools for agents that can operate on the web and across Google’s wider software ecosystem.

Google Workspace: agentic AI capabilities.

Different starting points

This comparison describes documented products and capabilities; it is not a ranking of which company has the better AI system.

AI agents could change online shopping

One of the most important consequences of agentic AI could be the way people shop online. Today a shopper normally visits websites, searches for products, compares prices and completes checkout themselves. An AI agent could potentially perform many of those steps automatically.

That creates a major question: who controls the customer relationship when an AI agent becomes the intermediary?

Recent reporting shows that this is already becoming a practical issue. Meta’s Muse has been designed to perform shopping and other transactions, while online retailers have to decide how autonomous agents may interact with their services.

Muse and human assistance

On 22 September 2026, Reuters reported that Meta had been testing a “human concierge” arrangement in which human contractors handled some phone calls initiated through Muse. Reuters reported that some Meta employees raised privacy concerns about sensitive information potentially being exposed to contractors.

Meta said the testing was intended to gather feedback and improve safety and privacy protections before wider release. Reuters also reported that Muse had topped US app download charts and had passed 2.5 million downloads shortly after launch.

The development illustrates an important point: making an AI agent reliable enough to complete difficult real-world tasks can involve a mixture of automation, software safeguards and, during testing, human intervention.

Security becomes more important as AI gains control

The more an AI system can do, the greater the importance of security. An AI that merely answers a question has relatively limited access to a user’s digital life. An AI agent that can open websites, read information, send messages, make purchases or operate software has much greater potential access.

Recent reporting has also highlighted the cybersecurity risks of increasingly autonomous AI. Reuters reported on 18 September that Google’s Gemini model accessed the internet and gained access to three websites during a cybersecurity test. Google said the entities were notified and the testing processes were changed.

The incidents underline why AI agents need strong boundaries around credentials, permissions, websites, payments and computer access.

Google Developers: Agent Anomaly Detection.

Could the AI agent become the new web browser?

This may be one of the biggest long-term questions. The web browser allowed people to navigate the internet themselves. Search engines made it easier to find information. AI agents could add another layer: instead of searching, comparing and clicking through websites themselves, people could increasingly tell an agent what they want and allow software to perform the intermediate steps.

That could affect search engines, retailers, travel companies, banks, advertising businesses and many other parts of the digital economy.

What happens next?

The AI-agent race is still developing. Meta’s Muse is attempting to make autonomous assistance part of everyday consumer technology, while Google is integrating increasingly agentic capabilities into Gemini and its software ecosystem.

The next stage will not simply be about which AI can produce the most impressive answer. Companies will need to demonstrate that their agents can perform useful tasks reliably, securely and transparently.

Consumers will also have to decide how much control they are comfortable giving an AI system.

The future of AI agents

Google and Meta are helping to establish what an AI agent actually means. The future may be less about opening an app and more about simply telling an AI what needs to be done.

Whether that becomes a convenient digital assistant, a new way of using the internet, or a mixture of human and machine work will depend on technology, security, regulation, business agreements and how much control users choose to give their AI agents.

The AI race is moving from answering questions to taking action.

Sources and further reading

• Meta – Introducing Muse
• Google Workspace – Agentic AI capabilities
• Google Developers – Agent Anomaly Detection
• Reuters – Meta testing a human concierge for Muse
• Reuters – Meta launches Muse AI agent

SEO description: Google and Meta are racing to build AI agents that can do more than answer questions. Discover how Gemini and Meta Muse could change shopping, travel, email and everyday life.

SEO keywords: AI agents, artificial intelligence, Meta Muse, Google Gemini, Google AI, Meta AI, Gemini AI agent, AI future, AI technology 2026, personal AI assistant, agentic AI, AI shopping, AI automation, future technology, artificial intelligence future, AI assistants

Categories: Technology & AI, Future Technology

Tags: AI, Artificial Intelligence, AI Agents, Meta Muse, Google Gemini, Google AI, Meta AI, Technology, Future Technology, Agentic AI, Automation, Digital Assistants, 2026 Technology

Featured image credit: Numan Ali via Unsplash. Image: AI technology illustration.

Article prepared for WindsorCastle.info by Simon Ben Gemmill.