{"slug": "the-part-of-the-internet-ai-agents-can-t-access", "title": "The part of the internet AI agents can't access", "summary": "A developer has launched Mobilerun, an open-source tool that gives AI agents a cloud-hosted Android phone so they can operate mobile-only apps that lack websites or public APIs. The agent reads the screen, decides what to tap, and verifies results, with live monitoring and human takeover for steps like login approvals. The developer argues browser agents like BrowserUse cover the web well but stall at the \"and now open the app\" step, and that real workflows will increasingly need both a browser and a phone.", "body_md": "Over the last year agents got pretty good at the web. They can open a site, read it, click through a checkout, fill out a form. Tools like BrowserUse made that a lot more accessible, and honestly I think that's great. For anything that lives in a browser, a browser agent is the right tool.\n\nThe funny part is that when you look at your own day, a lot of it doesn't happen in a browser at all. You approve a bank transfer in the banking app. You reply to a supplier on WhatsApp. You check a delivery in some carrier app that has a website from 2012, if it has one. Plenty of services today are mobile first, and some are mobile only. There is no website you could point a browser agent at, and there is no public API either.\n\nSo you end up with a strange situation. Your agent can write code, read your inbox and book a flight on a website, and the moment the next step is \"open the app\", it stops and hands the task back to you.\n\nThe obvious idea is to just give the agent a phone. That's basically what we do, but it's worth being honest about why it's not trivial.\n\nA phone is built for one person holding it, there is no clean way for software to say \"open this app, log in, go to the third tab, check the status\". The agent has to look at the screen, understand what's on it, decide what to tap and then check if it worked, the same way you do, just without getting bored.\n\nThen there's everything around it. Apps log you out, ask for a code, show a popup, update overnight and move the button you needed. A demo that works once on your desk is the easy part. Running the same task reliably, again and again, without someone babysitting it, is where most of the work actually is.\n\nAnd you don't want to do this on your personal phone. You want a phone the agent can use for its job, that you can watch, step into when needed and reset when you're done.\n\nMobilerun gives your agent a phone in the cloud. You hand it a task, it works through the actual apps on the device, and it comes back with the result. On the Android side it runs real Android apps on cloud phones, so the agent sees and uses the same app a person would use, including the ones that only exist on mobile.\n\nThe core is open source, so you can see how the agent reads the screen and decides what to do, and you can plug in the model and the agent you already use instead of switching to ours. You can also watch what it does live and take over when something needs a human, like a login approval, and then give control back.\n\nI don't think this is browser vs mobile. I mean... it's the same idea, an agent that uses software the way people do, just pointed at a different place. Browser Use and similar tools cover the web really well. Mobilerun covers the part that lives in apps. For a lot of real workflows you'll probably want both, the agent does the web part in a browser and the app part on a phone, and you only see the result.\n\nWhat I'm fairly sure about is that agents that stop at the browser will feel limited pretty soon. Not because the web isn't important, but because so much of people's actual digital life sits on their phone.\n\nIf you're building agents and keep hitting the \"and now open the app\" wall, give your agent a phone and see how far it gets. That's what we're building at mobilerun.ai.", "url": "https://wpnews.pro/news/the-part-of-the-internet-ai-agents-can-t-access", "canonical_source": "https://dev.to/priya_negi_9ffd29931ea408/the-part-of-the-internet-ai-agents-cant-access-2e0d", "published_at": "2026-11-16 09:12:07+00:00", "updated_at": "2026-11-16 09:21:12.914188+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "ai-products"], "entities": ["Mobilerun", "BrowserUse", "Android", "mobilerun.ai"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/the-part-of-the-internet-ai-agents-can-t-access", "markdown": "https://wpnews.pro/news/the-part-of-the-internet-ai-agents-can-t-access.md", "text": "https://wpnews.pro/news/the-part-of-the-internet-ai-agents-can-t-access.txt", "jsonld": "https://wpnews.pro/news/the-part-of-the-internet-ai-agents-can-t-access.jsonld"}}