The biggest challenge of developing WebBrain

One question we get asked frequently: what is the biggest challenge of developing WebBrain? The answer has two parts, and they compound each other.

The first part is model diversity. Unlike Claude for Chrome, which is tuned for one model and one browser, WebBrain has to make sure we work with all models — big and small, including tiny ones. That is the challenge on its own. Each LLM has its own quirks: some over-reason, some skip steps, some hallucinate CSS selectors that look plausible but do not exist. So we try to make it work with them all, in multiple browsers.

The second part is that websites evolve continuously. You cannot just write an exception rule for a site and expect it to hold for months. If you are lucky, it works for a few weeks before a redesign, a renamed class, or a new consent dialog breaks it. Any agent hard-coded for today's web is already aging.

That is the challenge: every model you add multiplies the quirks, and every week of uptime erodes the assumptions. It is a moving target on both axes.

But we love it. And we will continue.

Tags: #WebBrain #BrowserAgent #TinyModels #LLMQuirks #CrossBrowser #WebAutomation