Microsoft releases Webwright, an AI that writes browser automation code in the terminal

Microsoft Research has released an open-source framework called Webwright that lets AI models write Playwright code in the terminal, run bash commands, check logs, and fix errors to complete complex web tasks.

Unlike typical web agents that predict the next click or scroll from screenshots or DOM states, Webwright has the model write code directly. This makes it better for multi-step tasks like filling out forms or repeating actions across pages. The framework is simple, with only about 1000 lines of code.

In tests, Webwright with GPT-5.4 achieved 86.67% accuracy on Online-Mind2Web and scored 60.1% on Odysseys, a 35.1% improvement over the previous best result. This shows its potential for complex web automation.