One website, every screen size at once: our own Mac browser, from first day of work to a released beta in 12 days

The challenge

Every team that builds a website faces the same chore: a check that the pages look correct on desktop, tablet, and phone. Today that check has three methods. You drag a browser window narrower and wider. You switch between simulated devices, one at a time. Or you pay for a service that takes screenshots on remote computers. None of these methods shows all the target sizes live and side by side. None of them handles a site still in development on your own machine. The check-and-recheck loop costs time on every design change.

The existing tools ignore a second user completely: the AI agent. AI agents write more and more website code, but they have no reliable method for a check of their own work across screen sizes. The built-in tools serve a person at a keyboard. They do not serve software that must open a page, click through it, and collect screenshots alone.

We built VarioBrowser to close both gaps: one glance for a person, one set of commands for an AI agent.

What we built

VarioBrowser is a browser for the Mac. You open one website, and the app shows it at desktop, tablet, and phone sizes at the same time, plus any custom size. Each size is a real, live view on a single zoomable canvas. Sign in one time, and every screen size is signed in.

The views stay in sync. Scroll one view, and the others follow. Click a button in one view, and the app clicks the same button in the other views. Each size arranges the page differently, and the app still finds the correct button. Type into a form, and the text appears everywhere. Safeguards prevent an echo between the views and a duplicate submission.

Next to the app sits an interface for AI agents. Around 30 commands cover the page open, a click, text input, a screenshot, and diagnostic detail about the page state. The app installs this integration itself. An AI agent on the same Mac can therefore check a website at every screen size, with no setup by a person.

The paid tier adds three features. It simulates common vision deficiencies, in the live views and in a screenshot. An instant-replay recorder keeps the last moments on screen, so you save a video of a bug after the bug appears. A bug-context capture collects the page state, and it removes login tokens and other secrets automatically before the save.

Around the app we built an 8-page marketing website and a licensing system for accounts and paid tiers. We also did the release work for one product in two channels: an approval for the Mac App Store, and a direct download from our own site.

How we built it

Three decisions shaped the build.

A tight scope. We cut everything outside the core promise. The work therefore fit into 12 days.

AI agents as first-class users. The agent interface is not an add-on. An agent does everything the app itself does. That parity makes the agent use case real, not aspirational.

A complete release. Store approval, packaging, accounts, and licensing are the unglamorous final steps. They separate a demo from a product people install and pay for.

What this build demonstrates

VarioBrowser is our own product, and we built it the way we build for clients. Twelve days covered the full distance: a Mac app that works, an automation surface for AI agents, a marketing website, a licensing system, and a release through the Mac App Store plus a direct download. The product is currently available as a time-limited beta. We bring the same approach to a client engagement: a small scope, AI agents as real users, and release discipline.