Neon

I’ve been using it for a week Opera Neon and I don’t know if I’m testing the future of web browsing or participating in a psychological experiment on how much friction a human tolerates before returning to their usual browser. Probably both. Neon comes standard with everything that any veteran Opera user takes for granted: side messaging integrations, music apps in streamingthe multimedia panel… It is the reminder that, despite all the agentic experimentation, there is still an Opera underneath: practical, comfortable and designed for those who live glued to several platforms at the same time. The promise is seductive: a browser that not only answers questions, but act for you. Who browses, compares, reserves, creates. Who understands what you want to do and does it while you focus on more important things. Opera calls this “agentic AI“, and technically it is correct: Neon can take control of the browser, open tabs, fill out forms, compare products. It is AI with hands, it is Opera’s proposal for the same field as Perplexity with Comet or OpenAI with ChatGPT Atlas. The problem is that those hands are sometimes clumsy, unpredictable and dangerously overconfident. Opera Neon maintains all the classic features of Opera, such as the side panels to display messaging mini-applications or streaming music on an upper layer. In the image, Apple Music. Image: Xataka. Three brains in one body To understand Neon you have to accept that It is not an AI browser. It is a browser with three AIs living together. Chat, Do and Make. Each one with its function, its purpose, its personality. And here begins the first big problem: knowing which one to use at all times is a guessing exercise. Chat is the most familiar. A conversational chatbot that answers questions, summarizes pages, translates texts. Typical. It works well when you’re not making things up, which is about 70% of the time. I asked him to count the comments on several articles and he responded with 400 words explaining that there were none. when there were four. Do is where magic and terror live. You ask him to book a CrossFit class, find the cheapest flight to Lisbon, compare prices on headphones, unsubscribe from some newsletters. And sometimes it does. Open tabs, browse websites, fill out fields. Watching him work is hypnotic. It’s also slow, erratic, and occasionally catastrophic. In a test I asked her to add flowers to a store cart. Instead of somehow inferring my zip code or asking me about it, he directly introduced 28001: madridcentrismo to the song. While I, helpless, did click on the correct options that I was completely unaware of. There is no way to correct it while working. You can only watch, like someone who sees their autonomous car getting dangerously close to the cliff. A zip code just because, 350 km from my house. Image: Xataka. Neon spent an absurd amount of time wandering around the web, adding the bouquet to the cart, getting stuck on the shipping zip code, not feeling like anything productive was happening. Image: Xataka. Another example with Do: Image: Xataka. What he did was open Google Shopping, enter the term and not be able to click ‘Search’, apparently due to some subtle change in the website’s code. I gave it myself and Neon continued. It took a long time just to choose the order by price from lowest to highest. Finally he wrote the answer: Image: Xataka. Happy ending, although it is difficult to think of use scenarios where the use really compensates for the time and supervision it requires. If someone doesn’t know about Google Shopping, this is a good use case. If someone knows Google Shopping, they only have to do two clicks. Another example: reading some recipes Straight to the PalateI asked him to add all the ingredients necessary to make them to the Mercadona cart. Let’s go to trouble. Image: Xataka. Image: Xataka. This was one of those scenarios where there was no way I was going to complete the mission. Image: Xataka. Make is the most ambitious. Generate code, build web applications, create videos. I asked him for a memory game with Spanish vocabulary and he did it in minutes. Rough, but functional. It’s like having a mini-developer living in your browser, working in a virtual environment that disappears when you close the tab. A brilliant idea. A little polished perhaps, but brilliant. Image: Xataka. There are also the cardsa kind of templates prompts that function as mental shortcuts. You can combine them – “summarize + compare”, “decisions + follow-up” – or create your own so you don’t start from scratch every time you talk to the AI. It’s a simple but powerful idea: it makes user learning part of the system. Similar to what you propose Day with his Skills. It’s a good idea. What is not being said about Opera Neon Here comes the part that interests me the most, the one I read between the lines after a week living with this thing. Opera Neon is not really a product. It is a testing ground with product pricing. It is a public beta disguised as a premium service. And that wouldn’t be so much of a problem if it didn’t cost $20 a month. Let me be clear: I’ve seen enough technology launches to recognize when a company is testing concepts in the open field. And Neon is that. The bugs They are not occasional, they are structural, like the hallucinations. The Do agent disconnects if your computer goes to sleep. Chat responses are verbose. The Cards interface—those shortcuts prompts reusable—is full of examples with no real useful content. Cards examples interface. Image: Xataka. But there’s something more interesting going on here. Opera is making a counterintuitive bet at the worst possible time. We are in 2025: Google gives away Gemini in Chrome. Perplexity has Comet. The Browser Company (Arc’s company) has Day. Microsoft puts Copilot everywhere. And OpenAI recently launched ChatGPT Atlas. … Read more