The Future of Front Ends

With the invention of large language models, how might we witness the change in the World Wide Web's landscape?

We can broadly divide the World Wide Web into two use cases: consuming information and Providing information. Websites like Yahoo Finance, Y Combinator, and news sites will probably remain the same because they are intended to consume information. In the best case, they will just be input for AI agents to read and summarize users' choices on the daily feed.

However, AI agents have a huge opening to disrupt the World Wide Web in terms of providing information. Examples of those use cases are form-based websites, like applying for insurance, applying for a loan, creating an account, and filing a complaint. Currently, providing the information is a painful process. Sure, Chrome browsers and auto-filling forms have made a huge difference in filling out forms quickly, but it is still very manual and does not always work properly. These use cases can be replaced by an agent on the web where we type in answers to its questions. Could you have an AI agent ask you a series of questions? Sure, it's 100% Possible, but it's more annoying than filling out forms. Maybe it can be done via voice over a phone or computer. It would be tolerable but would still require a lot of work from the user.

What if we had an AI agent who did this on behalf of us? What if every user had his own agent? What would this agent be able to do? Maybe this agent can talk on your behalf? If the agent has all the information about you, it would just ask you for permission to talk to a friendly agent on a website, and then these agents can converse and accomplish the task you set out to do. All you did was click one button.

Let's say Google enables us to create our own agent. They already have much of our information, and creating an agent that knows about us is trivial. They will introduce a protocol called “Open Agent,” which enables other agents to talk to them [This is strictly unnecessary because LLMs are already good at conversing, but it may be needed for financial transactions or such]. Commercial websites can implement this protocol so their agent and your personal Google agent can talk to each other. For example, when you go to a car insurance website, it will say we support “Google Agents.” The friendly agent on the site will ask if she can talk to your agent. You will say yes, giving your agent permission to talk to it. In rare cases, if information is missing, your agent can ask you to input it. They talk, and in the end, you are given a Dollar amount. All you did was click a button, permitting them to talk.

If needed, you can probably add more information, like your SSN and income. You can permit it to read your email. Did you just buy a car? Your agent will know about it if you choose to do so.

Of course, this is just step one. We can already see the future, where the agent will probably be a step ahead of us. For example, the agent might already have the insurance quotes ready in anticipation of you buying the car; that's a step too far. We will only get comfortable with agents being in the loop.