Bill Gates expects AI agents to reshape the way people interact with computers. In a recent blog post, he describes software that could understand requests in everyday language, draw on information users choose to share, and handle tasks across areas of daily life.
The idea could change both how people use software and how it is made. Gates sees potential applications in health, education, productivity, entertainment, and shopping, but says technical and privacy challenges still need attention.
From separate apps to personal assistants
AI agents have been discussed for nearly 30 years, Gates notes, but recent advances have made them practical. Rather than opening a different application for every task, users could tell a device what they want in ordinary language and let an agent respond.
That response could be shaped by how much information a person chooses to share. Gates imagines agents building a detailed understanding of users’ lives so they can offer personalized answers and make proactive suggestions. The goal is a personal assistant for anyone who is online, with abilities beyond today’s technology.
This would shift the role of software from a collection of tools people operate one by one to an intermediary that interprets a request and acts on it. The usefulness of that approach would depend on the agent understanding what a person means and having enough relevant context to help.
Services could reach more people
Gates argues that AI agents could make services more affordable and accessible. In health care, he expects them to assist with primary care, health advice, and treatment decisions, while also helping medical staff make decisions and work more productively.
He also sees a possible role in mental health, where agents could make therapy more accessible. The source notes that research suggests large language models can describe emotions in more detail than humans can. Whether that makes machines good therapists remains an open question.
In education, agents could tailor learning to individual students and help teachers spend less time on paperwork and other tasks. Gates envisions systems that assess the effectiveness of one-on-one instruction and provide additional guidance to students. In this view, agents support teachers’ work rather than replace it.
Work, shopping, and entertainment
For productivity, Gates expects agents to take on tasks more independently, much like personal assistants. Examples he gives include preparing business plans and presentations and creating product images. He points to Microsoft’s Copilot and Google’s Assistant with Bard as tools already moving in this direction.
Agents could also help people choose and carry out purchases. Gates expects them to compare prices and services, recommend options, and tailor news and entertainment to individual interests. That puts the agent between a user and a range of services, helping make choices as well as respond to instructions.
The common thread is a move toward software that can act across different tasks. If that works as Gates describes, users may spend less time navigating individual apps and more time explaining what outcome they want.
A new platform, with hurdles to clear
Gates predicts agents will become the next computing platform, changing how software is used and written. He sees a future in which people could create apps and services without knowing how to code or design graphics: they would describe what they wanted, and an agent could write the code, design the app, create a logo, and publish it to an online store.
He points to OpenAI’s GPTs, released a few days ago, as an early sign of that possibility. In his view, agents could eventually help with a broad range of activities, with consequences for the software business and society.
Several obstacles remain. Gates identifies the need for new databases and structures that can track relationships and interests in nuanced ways while preserving privacy. Other open questions include how agents interact with people and one another, how to keep them aligned, how to lower generative AI’s high cost, how to make prompting easier, and how to reduce hallucinations.
Those problems leave important questions about reliability, privacy, and appropriate use. Gates says the technology industry and lawmakers will need to address them so agents can be used efficiently and appropriately. He has also said he does not expect a big leap from GPT-4 to GPT-5, while suggesting that fundamental problems such as hallucinations could be solved in the next few years.