Personal AI Agents Will Reshape the Internet

In the agent-centric internet era, users will directly interact with only a tiny part of the internet, while agents, like Jarvis, will handle most tasks by following human commands. Instead of users directly accessing websites, agents will perform most internet activities.

Humans will use only a fraction of the internet directly, with agents handling most command executions. This is likened to Iron Man's Jarvis, which operates based on the user's instructions.

These agents are developed by specific companies or open-source projects and provided to users in various forms, each with different functions and advantages.

One of the most important virtues of an agent is 'proactiveness,' which means the ability to perform tasks independently without being instructed by the user.

The key is to handle necessary tasks autonomously without unduly disturbing the user. This is similar to the qualities expected of an excellent secretary or subordinate.

Instinct analyzes the user's email content and independently suggests lecture drafts and anticipated Q&A. If the user links their email, the agent automatically helps with lecture preparation.

When provided with existing lecture material links, Instinct reflected this and automatically generated slides with 80% completeness.

Thus, agents find relevant information and proactively perform tasks, providing convenience, even without explicit user instructions.

Meta Muse independently identifies tasks that users might easily miss. Muse scanned past emails and notified the user that there were photo materials yet to be sent for promotional production.

This function helps prevent missed tasks by checking and presenting necessary work independently, even without explicit user instructions from the user.

By giving a single command to cancel a subscription, complex procedures are completed, or when booking a hotel, it compares points and cash payment options, reducing the hassle of web tasks. For example, complex processes like canceling a TV subscription are resolved instantly with one command.

Furthermore, for hotel bookings, agents perform tasks from checking dates to comparing point and cash payment availability, enhancing user convenience.

AI agents compare the efficiency of point versus cash payments for Marriott-affiliated hotels and recommend the optimal payment method to the user.

They even check late check-out availability by analyzing reviews and notices, and send direct inquiries to hotels based on the English email draft written by the user.

Thus, after the introduction of AI agents, the convenience of simple repetitive tasks is greatly felt, and the advantage of AI handling complex processes on behalf of users becomes significant.

After experiencing AI agents, users re-realize how cumbersome existing app-based internet usage, such as movie and restaurant bookings, was.

As complex website payment processes are simplified by agents, it feels difficult to return to past manual internet usage methods.

In the PC era, internet access and device usage restrictions made internet use a very cumbersome activity. One had to sit in front of a computer at home to use the internet.

With the advent of the mobile era, location restrictions were eliminated, leading to an explosive increase in internet usage. However, complex service structures and difficulties in user experience still existed.

In the agent era, the 'tediousness' of internet use is resolved as all purposes can be achieved through natural language commands instead of directly navigating to websites.

Internet usage can be broadly divided into two types. One is 'purpose-oriented' internet use, centered on clear tasks such as shopping, travel booking, stock trading, and information acquisition.

The other is 'non-purpose-oriented' internet use, simply for enjoyment, such as Instagram, YouTube, and TikTok. AI agents primarily maximize the efficiency of purpose-oriented internet use.

While there might be some imperfections currently, within 3 to 6 months, agents will become better at using computers than humans.

For clear-purpose activities such as search, shopping, and booking, agents will perform these directly, anticipating significant changes in the business models of existing companies like Booking.com, Hotels.com, Expedia, Google, and Amazon.

The maximization of agents' learning abilities and computer utilization capabilities will be a crucial factor in bringing about a major reorganization of the internet environment.

As AI generates fake traffic, the value of pure traffic generated directly by humans is expected to become even more precious. Accordingly, community sites will emerge as valuable platforms in the new era.

While major AI companies like OpenAI, Anthropic, and Google have not yet officially launched personal agents, Meta's 'Muse' currently holds the most advantageous position in terms of download performance.

Although the ultimate winner remains unknown, Meta's early success and stock price increase indicate its potential in this field. However, it remains to be seen whether it will be a short-term craze like Clubhouse or if it will endure.

Google is considered to be in the most advantageous position in the AI token and related chicken game. Google possesses the best capabilities among Silicon Valley companies in terms of AI optimization and cost-effectiveness.

With a full stack from TPU (Tensor Processing Unit) hardware to software, it can run AI much more cost-effectively. This, combined with massive core business cash flow, further strengthens Google's AI competitiveness.