Bill Gates: Agents Will Be AI's Biggest Race

On November 9, Bill Gates published an essay on his personal website titled "AI is about to completely change how you use computers," explaining how agents will disrupt the software industry in the coming years. As Oasis Capital shared in 42章经's "The People Betting Biggest on AI": we believe the core of this AI revolution lies in agents — large language models are merely one piece of infrastructure, and the ultimate goal is to generalize and universalize the expertise of professionals through agents. This is the most critical focus of this wave. OpenAI's product launch and Bill Gates's essay precisely confirm this macro trend, and we hope entrepreneurs will concentrate their efforts and invest more time and energy in this direction. Below is the full translation. Enjoy.

Image source: GatesNotes

Since Paul Allen and I founded Microsoft, my love for software has never faded. Yet despite decades of progress, software remains remarkably "clumsy" in many ways. To accomplish any task on a computer, you must specify which application to use. While you can draft a business proposal in Microsoft Word or Google Docs, these tools cannot send emails, share selfies, analyze data, schedule gatherings, or buy movie tickets for you. Even the best websites understand only fragments of your work, personal life, interests, and relationships, and their ability to use that information to serve you is limited. This kind of work is currently reserved for humans — your close friends or personal assistants.

But in the next five years, all of this will change completely. You will no longer need to switch between different apps for different tasks. You will simply tell your device what you want to do in plain language, and because it has developed a deep understanding of your life, software will deliver personalized responses based on how much information you are willing to share. In the near future, anyone with internet access will have access to a personal assistant powered by advanced AI.

This type of software — capable of understanding natural language and completing diverse tasks based on its knowledge of the user — is called an "agent." I have been thinking about agents for nearly 30 years, and mentioned them in my 1995 book The Road Ahead, but only recently, thanks to advances in AI, have they become truly practical.

Agents will not only change how people interact with computers — they will upend the software industry, triggering the biggest revolution in computing since we moved from typing commands to clicking icons.

A Personal Assistant for Everyone

Critics will point out that software companies have offered similar services before, without users fully embracing them. (People still joke about Clippy, the digital assistant we once included in Microsoft Office before abandoning it.) So why would people use agents? The answer: agents will be dramatically better. You will be able to have nuanced conversations with them. They will be far more personalized, and capable of much more than relatively simple tasks like writing letters. The gap between Clippy and an agent is as wide as the gap between a rotary phone and a smartphone.

Agents will be able to help with all your activities if you allow them to. With permission, they can track your online interactions and real-world locations to develop a deep understanding of the people, places, and activities in your life. They will grasp your personal and work relationships, hobbies, preferences, and schedule. You choose when and where you want them to step in to help you make decisions or complete tasks.

To understand how transformative agents will be, compare them to today's AI tools. Most of these are bots, confined to a single app, and typically only step in when you specifically ask them to. Because they cannot remember how you used them from one session to the next, they cannot learn and adapt to your preferences.

"Clippy was a bot, not an agent."

Agents are smarter. They are proactive — capable of making suggestions before you ask. They accomplish tasks across applications, and they improve over time because they remember your activities and recognize intentions and patterns in your behavior. Based on this information, they offer what they think you need, though you always remain in control of the final decision.

Imagine you want to plan a trip. A travel bot might only find hotels within your budget. An agent, knowing what time of year you'll be traveling and whether you always try new destinations or prefer returning to the same places, will suggest destinations accordingly. When asked, it will recommend activities based on your interests and appetite for adventure, and reserve tables at your favorite restaurants. If you want this kind of deeply personalized planning today, you need to pay a travel agent and spend time telling them what you want.

The most exciting impact of AI agents is that they will democratize services that are currently too expensive for most people. They will have an enormous impact across four key areas: healthcare, education, productivity, and entertainment and shopping.

Healthcare

In healthcare today, AI mainly assists with administrative tasks. For example, systems like Abridge, Nuance DAX, and Nabla Copilot can record audio during patient visits and organize it into notes for doctors to review.

The real transformation will come when agents can help patients with initial triage, provide advice on handling health problems, and determine whether you need to see a doctor. These agents will also help healthcare workers make smarter decisions and work more efficiently. (For instance, an app like Glass Health can analyze a patient summary and suggest possible diagnoses for the doctor to consider.) Helping both patients and healthcare workers will be especially important for people in poor countries, where many never get to see a doctor at all.

Clinical agents will roll out more slowly than other types because getting things right is literally a matter of life and death. People will need to be convinced that, despite their imperfections and potential for mistakes, health agents are beneficial overall. Of course, humans make mistakes too, and lack of access to medical care is itself a problem.

Mental health care is another service where agents will become accessible to virtually everyone. Today, weekly therapy sessions seem like a luxury. Much need goes unmet, and many who could benefit from treatment never receive it. A RAND study found, for example, that half of U.S. veterans who need mental health care don't receive it.

AI agents trained in mental health will make therapy more affordable and easier to access. Wysa and Youper are two early chatbot examples in this space. But agents will go much deeper. If you are willing to share sufficient information with a mental health agent, it will understand your life history and relationships. It will be available whenever you need it, and will never lose patience. With your permission, it could even monitor your physiological responses during therapy through your smartwatch — noticing when your heart rate spikes as you talk about problems with your boss, for instance — and advise when you should seek help from a human therapist.

Education

For decades, I have been excited about how software will simplify teachers' work and help students learn. It won't replace teachers — it will supplement their work by personalizing assignments for students, and freeing teachers from paperwork and other tasks so they can spend more time on the most important parts of their jobs. These changes are finally beginning to happen in dramatic ways.

The cutting-edge technology today is Khanmigo, a text-based bot developed by Khan Academy. It can tutor students in math, science, and the humanities — for example, explaining quadratic equations and creating math problems for practice. It can also help teachers with tasks like lesson planning. I have long been a supporter of Sal Khan's work and recently had him on my podcast to discuss education and AI.

But text-based bots are just the first step — agents will open up many more learning opportunities.

For example, few families can afford one-on-one tutoring to supplement classroom instruction. If agents can effectively capture what makes a tutor effective, they will make supplementary instruction available to everyone who needs it. If a tutoring agent knows a child loves Minecraft and Taylor Swift, it will use Minecraft to teach calculating volume and area, and use Taylor's lyrics to teach about narrative and rhyme schemes. Compared to today's text-based tutoring, this experience will be far richer — incorporating graphics and sound — and far more personalized.

Productivity

This is already a fiercely competitive space. Microsoft is integrating its Copilot feature into Word, Excel, Outlook, and other services. Google is doing similar things with Assistant, Bard, and its productivity tools. These copilots can do plenty: convert written documents into presentations, answer spreadsheet questions in natural language, and summarize email threads while representing each person's viewpoint.

Agents will be far more capable. Having an agent will be like having a person dedicated to helping you with various tasks and working independently on your behalf. If you have a business idea, for example, an agent can help you write a business plan, create a presentation, and even generate images of what your product might look like. Companies will make agents available to employees, allowing them to participate directly in meetings and answer questions.

Whether or not you work in an office, your agent will be able to assist you in the same way that personal assistants support executives today. If your friend has recently had surgery, your agent will suggest sending flowers and can handle the order for you. If you want to reconnect with a college roommate, it will coordinate with their agent to find a time to meet, and remind you before you arrive that their oldest child just started at the local university.

Entertainment and Shopping

AI is already helping you pick out a new TV and recommending movies, books, shows, and podcasts. Similarly, a company I invested in recently launched Pix, which lets you ask questions (like "Which Robert Redford movies would I like? Where can I watch them?") and makes recommendations based on your past preferences. Spotify's AI-powered DJ doesn't just play songs based on your preferences — it talks to you and can even call you by name.

Agents won't just make recommendations — they will help you act on them. If you want to buy a camera, you can have your agent read all the reviews for you, summarize them, make a recommendation, and place the order once you've decided. If you tell your agent you want to watch Star Wars, it will know whether you're already subscribed to the right streaming service, and if not, it will offer to sign you up. If you're not sure what you want to watch, it will provide customized suggestions and then play whichever movie or show you choose.

You can also get customized news and entertainment content tailored to your interests. CurioAI, which creates custom podcasts on any topic you ask about, exemplifies this trend.

A Seismic Shift in the Tech Industry

In short, agents will be able to help with virtually any activity and every area of life. This will have profound implications for the software industry and society as a whole.

In the computing industry, we talk about platforms — the underlying technologies that apps and services are built on. Android, iOS, and Windows are all examples of platforms. Agents will be the next platform.

To create a new app or service, you won't need to know how to write code or do graphic design. You will simply tell your agent what you want. It will be able to write the code, design the look and feel of the app, create a logo, and publish the app to an online store. OpenAI's launch of GPTs this week offers a preview of a future where non-developers can easily create and share their own assistants.

Agents will affect how we use software and how it's written. They will replace search sites because they will be better at finding and summarizing information for you. They will replace many e-commerce sites because they will find the best price for you and won't be limited to just a few vendors. They will replace word processors, spreadsheets, and other productivity apps. Today's separate businesses — search ads, social networking with ads, shopping, productivity software — will become one business.

I don't think any single company will dominate the agents business — there will be many different AI engines available. Today, agents are embedded in other software like word processors and spreadsheets, but eventually they will operate on their own. While some agents will be free to use (and supported by advertising), I think you will need to pay for most of them, which means companies will have an incentive to make agents work on your behalf rather than on advertisers'. The number of companies that have started working on AI this year suggests that competition will be unusually fierce, which will make agents very affordable.

But before the sophisticated agents I have described become reality, we need to confront many questions about the technology and how we use it. I have written before about the issues raised by AI, so in this article I will focus on agents.

Technical Challenges

No one has figured out yet what the data structure for an agent should look like. To create personal agents, we need a new kind of database that can precisely capture the nuances of your interests and relationships while maintaining your privacy and enabling fast retrieval of this information. We are already seeing new ways of storing information, such as vector databases, that may be better suited to storing data generated by machine learning models.

Another open question is how many agents people will interact with. Will your personal agent be separate from your therapy agent and your math tutor? If so, when will you want them to work together, and when should they stay in their lanes?

How will you interact with your agent? Companies are exploring various options including apps, glasses, pendants, pins, and even holograms. All are possibilities, but I think the first big breakthrough in human-agent interaction will be earbuds. If your agent needs to reach you, it will speak to you through your earbuds or display information on your phone. ("Your flight is delayed. Do you want to wait, or should I rebook you?") If you wish, it could also monitor sound entering your ear and enhance your auditory experience by canceling background noise, amplifying speech that's hard to hear, or making heavily accented speech easier to understand.

There are other challenges. There is no standard protocol yet that allows agents to talk to each other. The cost of agents needs to come down so everyone can afford them. It needs to be easier to prompt agents correctly to get the right answer. We need to prevent hallucinations, especially in areas like health where accuracy is critical, and ensure that agents don't cause harm to humans because of biases. We don't want agents to do things they shouldn't. (Though I am less worried about malicious agents than I am about human criminals using agents for malicious purposes.)

Privacy and Other Big Questions

As these technologies come together, online privacy and security issues will become even more urgent than they are today. You will want to be able to decide what information your agent has access to, and be confident that your data is only shared with people and companies you choose.

But who owns the data you share with your agent, and how do you ensure it is being used appropriately? No one wants to start getting ads related to something they told their therapy agent. Can law enforcement use your agent as evidence against you? When will your agent refuse to do something that could be harmful to you or others? Who sets the values embedded within agents?

There is also the question of how much information agents should share. Suppose you want to see a friend: if your agent talks to their agent, you don't want it to say, "Oh, she's seeing other friends on Tuesday and isn't planning to include you." If your agent helps you write a work email, it needs to know not to use your personal information or proprietary data from a previous job.

Many of these issues are already top of mind for the tech industry and lawmakers. I recently participated in an AI forum organized by Sen. Chuck Schumer, with other tech leaders and many U.S. senators in attendance. We shared ideas about these and other issues, and discussed the need for lawmakers to adopt strong legislation.

But other questions won't be decided by companies and governments alone. For example, agents may change how we interact with friends and family. Today, you can show you care by remembering details about their lives — like their birthday. But when they know it was probably your agent that reminded you and handled sending flowers, will it still mean as much to them?

In the distant future, agents may even force humans to confront profound questions about purpose. Imagine agents become so good that everyone can enjoy a high quality of life without doing almost any work. In such a future, how will people spend their time? Will anyone still want to get an education when agents have all the answers? Can a safe and prosperous society be maintained when most people have lots of free time?

But we are a long way from that.

In the meantime, agents are quietly arriving.

In the next few years, they will utterly transform how we live, both online and off.