Yahoo is using AI to generate takeaways from this article. This means the info may not always match what's in the article. Reporting mistakes helps us improve the experience.
The Browser Company was founded in 2019, raking in more than $100 million in funding on its way to releasing the Arc Browser earlier this year. Arc is not the end of the company's plans -- a new video offers a glimpse of the upcoming AI-powered Dia browser. CEO Josh Miller shows how Dia brings together all the latest AI concepts to automate your browser. It all makes for a nice demo, but are you ready to trust your online life to a machine?
Dia is built around the premise that the proper place for AI is not a standalone app or a button in your standard browser. Instead, Miller contends that AI should be fully integrated with the browser -- a predictable thing for "The Browser Company" to conclude.
The pitch makes sense if you aim to plug AI into as many tasks as possible. Many of us live in our browsers, accessing communication, shopping, meetings, and more without ever opening another application. Dia will be able to understand the context of your requests because it has all your browsing data and open tabs, something Miller says will remain private on your machine. It uses this "memory" to take actions and generate text wherever you point the mouse.
The demo video, which is also framed as a recruiting tool, shows how Dia can use its large language model component to seamlessly take over text generation with quick actions like "write next line" and "give me an idea." Because Dia is the browser, it doesn't need an API to access things like your email or calendar because you're already using those in the browser. Thus, Dia can allegedly take action in these tabs without additional login steps.
Automations are also available with the browser's "self-driving" module. The demo shows Dia tasked with buying some items from Amazon, which it appears to do adequately. We don't know if it bought the best possible hammer for the task, but it bought one. Dia is also told to send a series of emails with varying information to a group of people. Miller shows this happening in real-time, but he says the goal is for Dia to operate without the user's constant oversight.
Some of these demos are quite impressive, offering a glimpse of a future in which you might have a real AI assistant that can handle complex tasks that would flummox current digital assistants. However, much of the legwork still relies on a large language model to spit out text. Even the most capable models in the world regularly get things wrong, hallucinating falsehoods and refusing to admit fault.
The efficiency of AI automation agents that can control your PC is also still questionable. For instance, Anthropic's computer use model can cost several dollars for just a few minutes of automation. The browser company hasn't offered any specifics on the models it will use or where they will run.
So, while it's nice to imagine a browser that can email a dozen people with vital details, the reality might be much more frustrating. Today, every LLM-based AI includes a disclaimer that you should check its work. Is it really a time-saver if you're constantly babysitting your AI browser to make sure its automation doesn't get you fired? We'll find out in early 2025 when Dia launches.