Portuguese 
English 
Spanish 
4 min of reading 
0 comments 
Microsoft Unveils Plans to Replace Traditional Navigation with Autonomous Artificial Intelligence Agents Performing Complex Online Tasks.
Microsoft wants to transform the way people navigate the internet. The company believes that artificial intelligence has the potential to change the web as profoundly as mobile phones changed the world in the last 20 years. However, this transformation faces technical and practical challenges.
Microsoft’s Proposal for AI on the Internet
Generative artificial intelligence, which creates content based on user commands, gained prominence at the end of 2022.
At that time, OpenAI, backed by Microsoft, launched ChatGPT, a chatbot capable of writing anything from short texts to full essays with just a simple command. A few years later, Microsoft is presenting more ambitious plans.
-
An appliance that many people leave on during summer nights can consume up to 634 kWh per month, add more than R$ 560 to the electricity bill, and still incur higher costs due to incorrect temperature settings and a dirty filter, technicians warn.
-
India tests compact solar desalination aiming for 10,000 liters of drinking water per day, uses solar heat to evaporate seawater in stages, and can reach islands, condominiums, and campuses after a pilot in the far south of the country.
-
New equipment for vehicles promises to revolutionize driver safety on Brazilian roads with groundbreaking intelligent anti-sleep seat technology.
-
Chinese surprise the world by making a historic 7,000-ton building “walk” 60 meters using giant robotic legs in an unbelievable operation
Now, the company wants to surpass ChatGPT and its competitors. The idea is to build a new foundation for the internet.
According to Microsoft, AI is evolving rapidly. Companies in the sector are investing billions of dollars in infrastructure and restructuring their teams to adapt to the new landscape.
As proof of this restructuring, Microsoft laid off over 6,000 employees a week before announcing its new strategy.
The Agentic Web: Autonomous Agents Instead of Clicks
Microsoft defines its proposal as an “open and agentic web.” In it, users would send AI agents to perform tasks automatically.
Today, most people interact with AI through Google searches or conversations with chatbots. For Microsoft, these forms are already outdated.
In the company’s view, autonomous agents would be responsible for complex tasks. For example, a person could ask their agent to plan a vacation.
The agent would research flights, hotels, and create the itinerary, all automatically. This type of work still requires manual effort from users.
During the Build conference held in Seattle, CEO Satya Nadella detailed the new structure. He demonstrated how agents can fix computer codes, create PowerPoint presentations, and organize expense reports. These functions are possible thanks to what Microsoft calls the “AI stack.”
How Microsoft’s AI Stack Works
This stack consists of several layers. At the base lies the data infrastructure supported by Microsoft.
Above it are the AI models and agents, which interact directly with this information. Other companies are also building agents and AI models that operate on these foundations.
One of the central elements of this architecture is the communication protocol between agents. Microsoft bets on what it calls the Context Model Protocol. Kevin Scott, the company’s chief technology officer, compared the protocol to HTTP, the standard that enabled the traditional web to function.
This new protocol allows AI agents to access websites, gather data, and communicate with other agents automatically. “It’s filling an incredibly important niche,” Scott stated during a speech on May 19.
The Competition Among Tech Giants
Microsoft is not alone in this race. Since the launch of ChatGPT, various competitors have emerged.
Companies like Google, Meta, and Elon Musk’s xAI are also looking to develop similar technologies. All are trying to transition from a click-based internet to an agent-driven internet.
Chirag Shah, a professor of information science at the University of Washington, sees obstacles in this path.
The World Wide Web Consortium, the neutral body responsible for internet standards since 1994, does not require anyone to follow Microsoft’s protocol. According to Shah, other companies may adopt alternatives.
“I don’t see this as a change in the web,” Shah stated. “I see it as a set of ideas that already has competition.” Shah has previously worked for Microsoft but has no current ties to the company.
Problems with AI Accuracy
One of the biggest challenges for the advancement of the agentic web is ensuring the accuracy of the information generated by the agents. Since the launch of ChatGPT, generative AI has already faced several failures.
Recently, Google’s AI Overview provided fictitious answers while attempting to explain nonsensical idioms.
During the same period, xAI’s Grok chatbot spread conspiracy theories on the social network X. The messages were unrelated to the original topics, and xAI claimed the problem was caused by unauthorized changes.
On May 20, a day after Microsoft’s announcement at Build, the 404 Media website revealed that the Chicago Sun-Times published an article generated by AI with references to books that did not exist, by completely made-up authors and citations.
Although humans also make mistakes on the internet, they better understand the consequences of those errors.
Shah warns that, in delicate situations, the errors of agents can be even more severe. He compares: “It’s like giving $100 to someone and asking them to go shopping for you. That is low risk, but think about health, legal issues, and financial decision-making. There are significant consequences for making mistakes there.“
Adoption of the New NLWeb Language
Another step in Microsoft’s strategy is the launch of NLWeb, a new software language project. Just as HTML structured the traditional web, NLWeb promises to create a smoother AI experience. Websites like Eventbrite and Shopify have already begun adopting the technology.
NLWeb enables developers to create commands in natural language, making it easier for average users. According to Shah, this language meets programmers’ long-standing desire: to simplify searches on websites using plain language.
Despite the technological advancement, widespread usage still encounters barriers. Just as self-driving cars are still limited to testing in restricted markets, the full adoption of the agentic web may take time. For Shah, the world is still not ready to relinquish its autonomy to these systems.
“We are not ready, all of us, to give up our autonomy in favor of these systems,” he concluded. “They will not radically replace the internet as we know it.”
Be the first to react!