The Oncoming Era of Agents

How AI Will Fundamentally Reshape Your Interaction with Technology And Revolutionize the Software Industry

Software has advanced significantly over the decades, yet in many fundamental ways, it remains surprisingly basic.

Currently, performing any task requires users to select a specific application. You might use one program for writing a business proposal, but it can't assist with sending emails, analyzing data, or planning social events. Even sophisticated platforms lack a holistic understanding of your professional and personal life, your interests, and your relationships, limiting their ability to act on your behalf in a meaningful way. This level of assistance is something only another person, like a personal assistant, can provide today.

This paradigm is expected to shift entirely within the next five years. Instead of juggling different apps, you will simply articulate your needs to your device using natural language. The software, depending on the permissions you grant, will respond in a highly personalized manner, drawing from a deep understanding of your life. In the near future, any online user will have access to an AI-powered personal assistant far more advanced than current technology.

This emerging category of software, known as an "agent," is defined by its ability to comprehend natural language and execute a wide array of tasks based on its knowledge of the user. The concept of agents has existed for decades, but recent breakthroughs in artificial intelligence are finally making them a practical reality.

Agents are poised to do more than just alter how people interact with computers; they are set to disrupt the entire software industry, heralding the most significant shift in computing since the move from command-line interfaces to graphical user interfaces.

A Personal Assistant for Everyone

Skeptics might note that software companies have previously introduced similar concepts, which failed to gain widespread adoption (consider the infamous "Clippy" digital assistant). Why will this time be different?

The key difference is that modern agents will be vastly superior. They will enable nuanced, contextual conversations and offer deep personalization, going far beyond simple tasks like drafting a letter. Comparing Clippy to a true AI agent is like comparing a rotary phone to a smartphone.

With user consent, an agent can assist with nearly every aspect of your life. By monitoring your digital interactions and physical locations, it can build a robust understanding of the people, places, and activities that matter to you. It will learn your professional and personal relationships, hobbies, preferences, and daily schedule, offering to intervene and assist at appropriate moments, while the user always retains final control.

An important distinction: Clippy was a bot, not an agent.

To appreciate the transformative potential of agents, it's useful to contrast them with today's AI tools, most of which are "bots." Bots are typically confined to a single application and react only when prompted with a specific keyword. They don't retain memory of past interactions, so they cannot learn your preferences or improve over time. Clippy exemplified this bot-like behavior.

Agents, in contrast, are far more intelligent. They are proactive, capable of offering suggestions before you even ask. They operate across multiple applications and continuously improve by remembering your activities and recognizing patterns in your behavior. Based on this learned context, they can anticipate your needs, though the user always has the final say.

Consider planning a trip. A standard travel bot might find hotels within your budget. An agent, however, would know your preferred travel seasons and whether you enjoy exploring new destinations or revisiting favorite spots, and could suggest locations accordingly. It would recommend activities aligned with your interests and risk tolerance and even book tables at restaurants you're likely to enjoy. Achieving this level of personalized planning today requires hiring a human travel agent and investing significant time in briefing them.

Perhaps the most significant impact of AI agents will be the democratization of services that are currently unaffordable for the majority. Their influence will be especially profound in four key areas: healthcare, education, productivity, and entertainment.

Healthcare

In healthcare today, AI primarily assists with administrative duties. For instance, services like Abridge, Nuance DAX, and Nabla Copilot can transcribe audio from a doctor's appointment and draft notes for the physician's review.

A more substantial transformation will occur when agents can assist patients with basic triage, offer advice on managing health issues, and help them determine if they need to seek professional treatment. These agents will also empower healthcare workers by aiding in their decision-making and boosting productivity. (Existing applications like Glass Health can already analyze a patient's case summary and suggest potential diagnoses for a doctor's consideration.) This support for both patients and clinicians will be particularly impactful in lower-income countries, where access to a doctor is often limited.

The rollout of these clinician-agents will likely be more cautious than others, as accuracy is a matter of life and death. The public will need to see clear evidence that health agents are beneficial overall, accepting that while they won't be flawless, they can improve upon a system where human error and lack of access are also significant problems.

A RAND study found that half of all U.S. military veterans who need mental health care do not receive it.

Mental health care is another field that agents will make accessible to nearly everyone. While weekly therapy might seem like a luxury today, there is a vast, unmet need for mental health support. For example, research has shown that a significant portion of U.S. military veterans in need of mental health services do not get them.

Well-trained mental health AI agents will make therapy more affordable and readily available. Early chatbots like Wysa and Youper are pioneering this space, but future agents will offer much deeper engagement. If you grant it permission, a mental health agent could understand your life history and relationships. It would be available on-demand, offering a patient and non-judgmental ear. With your consent, it could even monitor your physiological responses via a smartwatch—noticing if your heart rate spikes when discussing a stressful topic—and suggest when it might be beneficial to consult a human therapist.

Education

For years, there has been excitement about software's potential to ease the burden on teachers and enhance student learning. The goal is not to replace educators but to augment their work—personalizing lessons for students and freeing teachers from administrative tasks so they can focus on high-impact instruction. These long-awaited changes are finally materializing in a significant way.

A current leading example is Khanmigo, a text-based AI from Khan Academy. It can tutor students in subjects like math, science, and humanities, capable of explaining concepts like the quadratic formula and generating practice problems. It also serves as a tool for teachers, assisting with tasks like lesson planning.

However, text-based bots are just the beginning; agents will unlock a far wider range of learning opportunities.

For instance, very few families can afford private, one-on-one tutoring to supplement classroom instruction. If agents can successfully model the key attributes of an effective tutor, they can make this personalized support available to everyone. A tutoring agent that knows a child is interested in Minecraft and the music of Taylor Swift could use the game to teach concepts of volume and area, and use song lyrics to explore storytelling and rhyming structures. The experience will be far more dynamic—incorporating graphics and audio—and more customized than today's text-based tutors.

Productivity

Competition in this space is already fierce. Microsoft is integrating its Copilot across services like Word, Excel, and Outlook. Google is pursuing a similar strategy with Assistant with Bard and its suite of productivity tools. These "copilots" are highly capable, able to convert a text document into a slide presentation, answer natural language questions about spreadsheet data, and summarize lengthy email chains, accurately representing each participant's viewpoint.

Agents will push these capabilities even further. Having an agent will be akin to having a dedicated assistant who can manage various tasks for you, even working autonomously if you wish. If you have a business concept, an agent could help you draft a business plan, create a corresponding presentation, and even generate mock-up images of your product. Companies will be able to provide agents for their employees to consult with directly, even having them participate in meetings to answer questions on the fly.

If a friend recently had surgery, your agent could offer to send flowers and then handle the ordering process for you.

Regardless of your profession, your agent can provide the kind of support that personal assistants offer to top executives today. If a friend has a medical procedure, your agent could proactively suggest sending flowers and then manage the purchase. If you mention wanting to reconnect with an old college roommate, it could coordinate with their agent to find a suitable time. Just before your meeting, it might remind you that their eldest child recently started attending the local university.

Entertainment and Shopping

AI is already being used to help you choose a new television or recommend movies, books, and podcasts. For example, the service Likewise recently launched Pix, which allows you to ask conversational questions (e.g., "Which Robert Redford movies would I enjoy and where are they streaming?") and provides recommendations based on your viewing history. Spotify's AI DJ not only curates music based on your taste but also provides commentary in a personalized, conversational style.

Agents will go beyond mere recommendations and help you act on them. If you're looking to buy a camera, your agent could read and summarize all the relevant reviews, present a final recommendation, and place the order once you approve. If you tell your agent you want to watch Star Wars, it will know which streaming services you subscribe to and, if necessary, offer to sign you up for the right one. If you're unsure what you're in the mood for, it will provide tailored suggestions and then seamlessly play your chosen movie or show.

You will also be able to access news and entertainment customized to your specific interests. Services like CurioAI, which can generate a custom podcast on any topic you request, offer a preview of this future.

A Shock Wave in the Tech Industry

In essence, agents will be capable of assisting with nearly any activity in every facet of life. The consequences for both the software business and society will be immense.

Within the computing industry, we often speak of platforms—the foundational technologies upon which applications and services are built. Android, iOS, and Windows are all platforms. Agents represent the next major platform.

To create a new app or service, you'll just tell your agent what you want.

Creating a new application or service will no longer require knowledge of coding or graphic design. You will simply describe your needs to your agent. It will be able to write the necessary code, design the user interface, generate a logo, and even handle publishing the app to an online marketplace. The recent launch of custom GPTs by OpenAI provides a glimpse into a future where non-developers can effortlessly build and share their own assistants.

Agents will transform not only how we use software but also how it is created. They will likely supersede search engines by being more adept at finding and summarizing information. They will replace many e-commerce sites by finding you the absolute best price, unrestricted by vendor limitations. They will also absorb the functions of word processors, spreadsheets, and other productivity applications. Entirely separate business sectors today—such as search advertising, ad-supported social networking, online shopping, and productivity software—will likely converge into a single, integrated industry.

It's unlikely that a single company will dominate the agent market. Instead, a multitude of different AI engines will be available. While agents are currently embedded within other software, they will eventually operate as standalone entities. Although some agents may be free and ad-supported, it's probable that many will be subscription-based. This means companies will be incentivized to act in your best interest, not an advertiser's. Given the explosion of companies entering the AI space, intense competition will likely make agents highly affordable.

However, before the sophisticated agents described here become commonplace, several questions regarding the technology and its implementation must be addressed.

The Technical Challenges

No one has yet perfected the data structure for an agent. To build truly personal agents, a new type of database is needed—one that can capture the subtleties of your interests and relationships and recall information instantly while rigorously protecting your privacy. New data storage methods, like vector databases, are emerging and may be well-suited for storing data generated by machine learning models.

Another unresolved question is how many different agents a person will use. Will your general personal agent be distinct from your specialized therapy agent or your math tutor? If so, when should they collaborate, and when should they remain separate?

If your agent needs to check in with you, it will speak to you or show up on your phone.

How will you interact with your agent? Companies are exploring numerous possibilities, including apps, smart glasses, pendants, pins, and even holograms. While all of these are potential avenues, the first major breakthrough in human-agent interaction will likely be audio-based, perhaps through an earbud. Your agent could speak to you when needed ("Your flight is delayed. Would you like to wait, or shall I find a new booking?"). If you permit it, the device could monitor ambient sound, enhancing your hearing by filtering out background noise or clarifying speech that is difficult to understand.

Other challenges remain. There is not yet a universal protocol allowing agents to communicate with each other. The cost of the underlying computation must decrease to make agents accessible to everyone. We need to make it easier for users to prompt an agent effectively to get the desired result. We must prevent AI "hallucinations" (fabricated information), particularly in high-stakes areas like health, and ensure that agents are not perpetuating harmful biases. Finally, we must build safeguards to prevent agents from being used for malicious purposes.

Privacy and Other Big Questions

As this technology matures, issues of online privacy and security will become more critical than ever. You will need the ability to control precisely what information your agent can access, ensuring your data is shared only with the people and companies you explicitly authorize.

But crucial questions remain. Who owns the data you share with your agent, and how can you be certain it is used appropriately? Nobody wants to see advertisements related to a confidential conversation with their therapy agent. Can law enforcement subpoena your agent's data to be used as evidence against you? When should an agent refuse to perform a task that could be harmful? Who determines the ethical values embedded within these agents?

There is also the matter of how much information an agent should share. For example, if you want to meet a friend and your agent communicates with theirs, you wouldn't want it to disclose, "They are meeting other friends on Tuesday and decided not to include you." Similarly, if your agent assists with work emails, it must be trained not to use your personal information or proprietary data from a past employer.

Many of these concerns are already being debated by the tech industry and lawmakers. But some issues will fall outside the purview of companies and governments. For instance, agents could fundamentally alter how we interact with friends and family. Today, remembering a detail like a friend's birthday is a sign of caring. But when they know your agent likely reminded you and handled the flower delivery, will the gesture hold the same meaning?

In the distant future, agents might compel humanity to confront profound questions of purpose. Imagine a world where agents are so advanced that everyone can enjoy a high quality of life with minimal work. In such a future, how would people spend their time? Would anyone pursue an education when an agent holds all the answers? Can a society remain safe and vibrant when most of its population has abundant leisure time?

We are a long way from that reality. But in the meantime, agents are on the horizon. Within the next few years, they will begin to utterly transform our lives, both online and off.