Stagehand (Browserbase)

Stagehand sits in the awkward middle ground between raw Playwright and full-stack browser agents like Manus, and that's exactly where it becomes useful. I've used it to scrape product catalogues from sites that change their DOM structure weekly—places where XPath selectors rot and vision models overkill the problem. Stagehand's autonomy is narrower than you might expect: it doesn't plan multi-step research workflows or decide what to click next based on fuzzy goals. Instead, it gives you a Playwright-like API with built-in resilience. You still write the script, but Stagehand adapts to layout shifts, waits intelligently for dynamic content, and retries flaky interactions without you babysitting selectors. The sweet spot is bespoke agents where you know the rough shape of the task but the page structure isn't stable. I built a QA agent that logged into a staging environment, filled forms with edge-case data, and screenshotted error states. Stagehand handled the login flow even when the auth provider changed its button classes mid-sprint. Where it falls short: documentation assumes you're already comfortable with browser automation primitives. If you're coming from no-code tools, the learning curve is steep. It's also overkill for static scraping—just use Playwright. And if you need true autonomy ("go research competitors and summarise findings"), you want something higher-level. Browserbase's cloud infra is a natural pairing, but Stagehand runs locally too, which matters for dev iteration speed. The open-source repo is active, but the community is smaller than Playwright's, so expect to read source code when things break. I'd reach for Stagehand when I'm building a custom browser agent that needs to survive real-world DOM chaos without constant rewrites. It's not a product you deploy and forget—it's a framework for developers who want control but don't want to reimplement retry logic and element-wait heuristics from scratch.