Last month, Google DeepMind dropped Gemini 2.5 Computer Use, and I’ve been turning it over in my head ever since. Here’s the thing: this isn’t just another model update. It’s an AI that can actually navigate the web like a human would, clicking links, typing into forms, and moving through multi-step tasks without someone hovering over it.
New Feature / Update: Gemini 2.5 Computer Use
What Is It?
Gemini 2.5 Computer Use is a specialised model built on Gemini 2.5 Pro that lets AI agents interact directly with user interfaces. It’s available via the Gemini API for developers, and it works by combining visual understanding with reasoning to complete tasks autonomously. Think of it as an AI that can see what’s on your screen and actually do something about it.
Where previous models could only understand text or images in isolation, this one processes what it sees on a webpage or application and then performs actions. It can fill out forms, navigate between sites, extract data, and move through workflows that require multiple steps. Google’s already confirmed it outperforms alternatives on benchmarks for this kind of work.
Why Does It Matter?
Let me ground this in actual tasks, because that’s where it gets interesting.
For marketing and ops teams, imagine you’re pulling product data across five different retailer portals to build a comparison spreadsheet. Normally, someone sits with those windows open, copying and pasting for an hour. With Computer Use, you can set an agent loose to extract that data, format it, and dump it into a spreadsheet. Same task, fraction of the time. No more pixel tantrums on your end.
For developers and product teams, this becomes a bit of a mission in the best way. You can automate things like booking appointments, processing applications, or syncing data between systems that don’t have direct API connections. It’s particularly useful when you’re working with legacy systems or third-party platforms that weren’t built with automation in mind.
The practical bit: this got released in October 2025 and is available now via the Gemini API. If you’re already using Gemini in your stack, you can start experimenting with it.
What’s the Catch?
Speed is solid, but it’s still an agent doing visual interpretation and then executing. It’s not instant. It’s also working through a user interface rather than an API, so it’s only as reliable as the interface stays consistent. If a website redesigns its form layout, the agent might get confused mid-task. That said, the benchmarks show it handles this better than competing models.
The Bigger Picture
This is part of a broader shift happening right now. October saw a heap of automation updates across the industry. Salesforce finished acquiring Regrello for AI workflow automation. HubSpot released 100+ updates, including better automation overviews. Even Automation Anywhere’s sunsetting older tools and pushing toward document automation. The message is clear: we’re moving from simple if-this-then-that automations to actual autonomous agents that can think through multi-step processes.
Computer Use fits right into that. It’s one more tool in the kit for teams trying to do more without hiring more.
