The development of AI models for web navigation is rapidly advancing, with a focus on creating more efficient and capable agents to handle online tasks. Collaborations with entities like Google DeepMind have led to significant improvements in AI performance, particularly with the Gemini 2.5 Computer Use models, which excel in accuracy, speed, and cost efficiency. Through the use of the Browserbase infrastructure, the training and evaluation of these models have been drastically expedited, allowing for parallel processing and reduced runtime. However, challenges remain in testing these agents due to the dynamic nature of the web. To address this, efforts are being made to enhance transparency and standardization in evaluation processes, including the publication of extensive human-verified evaluation datasets. Open-source tools like the Stagehand Evals CLI facilitate broader participation in this development, inviting community contributions to improve benchmarks. As these technologies progress, the potential for AI to navigate and interact with the web alongside humans is becoming increasingly feasible, heralding a future where digital tasks can be automated across various interfaces.