In recent months, the market has seen a surge of browser agents that use computers just like humans do. It's no longer just about automating form filling; these agents are now performing complex tasks for personal and corporate assistants.
Major players like OpenAI (Operator), AnthropicAI (Claude Computer Use), and GoogleDeepMind (Project Mariner) are actively developing their directions. Open source frameworks like browser_use and Stagehanddev are gaining popularity on Github, racking up tens of thousands of stars.
Currently, the most progress is being made in vertical solutions: specialized agents for marketing, sales, QA, and HR (for example, Astral, Spur, Unify, SonicJobs). They operate more reliably due to their focus on narrow scenarios, unlike general-purpose models that often turn out to be "jack-of-all-trades, master of none."
Despite the rapid progress, agents are still far from automating truly valuable tasks. For instance, none have scored above 9.2% on the CUB benchmark (real workflows). Limitations include weak memory, unstable execution of long action chains, and coordination issues between different applications. Speed and accuracy remain a compromise, which is especially important for complex corporate scenarios.
Source: https://www.thetasoftware.ai/blog/the-browser-agent-landscape
