Webwright Review: Reusable Browser Automation Scripts

Quick answer

Webwright is Microsoft's open-source, code-as-action browser agent. It gives a coding model a terminal and browser, then requires the completed task to live in a rerunnable Python and Playwright script. It is a strong fit for long, repeatable browser workflows where you want inspectable code instead of opaque click-by-click automation. You still need a supported model backend or host agent and someone comfortable running Python automation.

A browser that remembers what it learned

Most AI browser tools work like someone with short-term memory loss. They look at the screen, guess the next click, and start over from scratch every single time. Ask them to fill out a form next week — they've forgotten everything.

Webwright, released by Microsoft Research, works differently. Instead of guessing click by click, it writes a small script the first time it figures something out. The next time you need the same task done, it just runs the script. No guessing. No starting over.

The results are striking. On a standard industry test for complex, multi-step web tasks, it scored 60% — compared to 33% for the same AI without this approach. That's nearly double.

What makes it interesting for a business owner: imagine you have a repetitive web task — pulling competitor prices, filling a supplier form, checking inventory across three sites. Today, you either do it manually or pay someone to do it. With an approach like this, you solve it once and it becomes a little tool you own. Reusable. Shareable. Yours.

It's open source and small enough that a developer can understand the whole thing in an afternoon — which means it's actually possible to build something specific to your business on top of it.

Words worth knowing

Browser agent — an AI that controls a web browser the way a person would: clicking, typing, navigating. No human needed.

Script — a small saved set of instructions. Like a recipe a computer can follow on its own.

Open source — the code is public and free to use, modify, or build on.

Benchmark — a standard test used to compare how well different AI systems perform on the same task.

If you have a web task your team does every week, it's worth asking: could this be the last time someone does it by hand?