Opera’s Neon Shows How Confusing AI Browsers Still Are

The trick to understanding Opera’s Neon Browser is recognizing that it’s not just a browser with an AI bot added to it, but a browser with three AI bots all living side by side. This is both a strength and a weakness, because while you don’t have to leave to do all the AI ​​things you want, knowing where to go for which AI tasks can be really confusing.

Opera began removing people from the waitlist for its AI browser, Neon, last month. It enters an increasingly crowded market of AI-powered browsers, including Google’s Gemini-infused Chrome, Perplexity’s Comet, and The Browser Company’s Dia. Unlike many of these competitors, Opera charges $19.90 per month, setting high standards for a product that most people get for free.

Neon has a similar configuration to other Opera browsers. It has features like a built-in ad blocker and VPN, as well as a sidebar that you can populate with apps like WhatsApp and Facebook Messenger. AI features are brought to the forefront on the home and new tab screens. Below the search bar is a toggle with four settings: regular Internet search; a chatbot called Chat; a browser control agent called Do; and an AI construction agent called Make. (Opera says Neon is built on AI models from OpenAI and Google, but it doesn’t specify which model is used for which part of Neon.)

Chat is the simplest feature, an in-browser AI assistant that will likely be familiar to users of any AI chatbot by now. It is accessible at the start of a search query, but is also located in the upper right corner of the browser for access at any time. Chat easily responded to our quick search queries and could answer specific questions on the page we were browsing, such as our request for a summary of the latest research in quantum computing. But it wasn’t perfect. The chat responses were so wordy that we felt like we had to do some real work to overcome them. When we asked Chat to tally and summarize comments from five of the most recent Edge stories, he responded with about 400 words to tell us there weren’t any.

It was this exercise that showed us that Chat couldn’t always read the Internet, but he made it look like he totally could. Another time it was said that there were no comments out of three Edge stories when there were actually four. We found it odd that Chat instead offered an estimate of what “first comments on tech news sites” often include.

Opera’s executive vice president for browsers, Krystian Kolondra, said the bot failed because we chose the wrong tool. The chat managed to note the number of comments listed at the top of the Edge articles when Opera tried it, Kolondra said. However, summarizing the comments requires expanding the comments section on the web page, which requires a click. Instead, we should have sent our requests to the AI ​​Do agent, Chat’s more active cousin that takes over the browser and performs tasks for you.

We tried Do on a number of tasks: booking a CrossFit class, booking a massage for less than $50 at a nearby spa, and finding PDFs of sewing patterns for baby rompers. While Do was running, we couldn’t return to chat in the same window if we wanted to ask follow-up questions about the task at hand. There is also no way to correct Do while it is in action. We watched in horror as the robot scrolled past perfectly pretty floral arrangements we’d tasked it with finding for a friend, only to add a monstrous funeral wreath to our cart, even as we clicked through the best options. Another time, Do said there were no theater tickets for a January show when a cursory check showed plenty. It’s hard to trust something after such unwavering but misplaced trust.

As with other AI browsers, doing things with Do was also slower than doing it ourselves, although it hinted at what outsourcing the general mundanity of web browsing might be like. And using Do doesn’t mean you can still do a full check. Sometimes he encountered obstacles that only a human could overcome. When this happened, the Do tab at the top of the screen flashed an easily missed shade of red, telling us that we needed to step in and help the robot on its way.

In addition to Chat and Do, there is also the AI ​​Make agent, which can create small web tools for you. Make exists in a virtual computer where it downloads the software, scripts, or in our case, images needed for your creation without cluttering up your home computer. We asked for a simple memory game with an introduction to Spanish vocabulary. It worked, and within minutes we matched the word “book” to the picture of a book. The game was clunky, but it was convenient to close the tab and know that all the images in the book disappeared with it.

The final selling point of Opera’s Neon is the cards, effectively pre-written prompts that can be used on any of the AI ​​agents that Opera says act as “power-ups” for your interactions with the AI. Cards could, in theory, save us from having to write a prompt, but right now it’s hard to see ourselves needing the ability to reuse a set of instructions. The App Store-like interface is largely filled with content from the Neon team, ranging from whimsical prompts that rewrite sites as if Yoda spoke them to more serious prompts for news aggregators. Opera hopes the platform will fill with useful user creations as more people use it, but there are few of those today.

Sometimes using Neon felt a bit like working with a hapless intern we never asked for, rather than working with sophisticated technology that saved us time. Often, one of its AI systems would ask for feedback and then simply launch into a task without waiting for a response. Given its ability to use the browser, it’s all too easy to imagine where this proactivity could go wrong, like sending a slew of LinkedIn queries to people you just wanted to anonymously stalk in a professional capacity. One time we responded letting him know everything was fine and we were moving forward, and Neon said “I’m glad you think so!” and immediately stopped working. Kolondra told us that Neon will pause to respond to requested comments in the future, but that the feature is disabled because it’s not ready yet.

Kolondra acknowledged that Neon is currently a work in progress. “In general, Opera Neon is in the early access release phase and is being made available to people who want to participate in the development of this product,” Kolondra said. The edge.

But Neon is also a paid subscription product, asking $20 per month, for something that’s widely available for free elsewhere. Right now, that’s a tough sell – especially since Neon seems more like an AI browser that we have to adapt to than a browser that’s smart enough to adapt to us.