Opera's new AI operator wants to do your online chores

Opera’s AI agent browses the web using code, not screenshots.

While most browser automation tools still feel like watching a robot clumsily peek at your screen, Opera took the wraps off its new AI Browser Operator on March 3. Instead of streaming your display to a remote server or clicking pixels, this built-in agent runs locally on your machine and reads the actual DOM tree behind the scenes.

Bypassing pixels for pure HTML

Most current attempts at web agents rely on capturing screenshots, shipping them off to cloud-hosted vision models, and guessing where to move a virtual mouse. It is sluggish, expensive, and breaks whenever a cookie banner pops up.

Opera is taking advantage of building browsers for decades. The Browser Operator reads the page’s underlying code and layout structure directly on your device.

Because it consumes text rather than pixels, it can digest an entire page in one glance without scrolling down. Annoying verification or cookie consent popups do not even trip it up, because the agent can simply interact with background elements that are technically hidden from human eyes.

More importantly, you stay logged in securely. Because the AI runs inside your client, your active cookies, credentials, and browsing history stay right on your laptop instead of being mirrored on an external virtual machine.

Knowing when to hand back the steering wheel

Handing full agency to an AI is terrifying if it has access to your wallet. Opera built the agent around a strict human-in-the-loop setup.

You launch it from the sidebar or the command line to handle repetitive tasks like buying everyday items, booking tickets, or scraping web snippets into documents. When it reaches a checkout page or asks for card numbers and passwords, the agent pauses completely.

During those moments, Opera says no keystrokes or inputs are processed by the AI or shipped to external servers. You fill in the payment details yourself, hit resume, and the agent takes back over to finish the job. If you spot a mistake, you can cancel at any second or simply instruct it to cancel the accidental order.

Why it matters

This is one of the smartest architectural choices we have seen for web agents so far.

By leveraging DOM text and local browser context instead of brute-force computer vision, Opera avoids the latency and cloud inference costs that plague competitors. You do not have to worry about handing your Amazon session tokens to an unvetted third-party server just to automate a purchase.

For anyone tired of filling out repetitive forms or compiling web data manually, this makes agentic browsing genuinely appealing. We think this local, code-first approach makes a lot more sense than waiting ten seconds for a vision model to find a "checkout" button.

That said, modern web engineering is notoriously messy. Heavily obfuscated single-page apps, dynamic shadow DOMs, and aggressive bot-detection systems frequently break non-standard parsers. Opera still has to prove that Browser Operator can handle hostile or poorly coded websites without getting stuck.