AI Agent Manus Achieves 86.5% on Gaia Benchmark, Bolstering Garry Tan's Vision for Human-like Web Agents

Garry Tan, the influential CEO of Y Combinator, recently articulated a bold vision for the future of artificial intelligence, asserting that web agents should possess the capability to execute user commands with human-like autonomy. His statement, made on September 3, 2025, underscores a significant shift in expectations for AI, moving beyond mere information processing towards active, intelligent task fulfillment. This perspective aligns with ongoing advancements in AI agent technology, which aim to replicate human decision-making and interaction patterns within digital environments.

In a social media post, Tan explicitly stated, > "Web agents should be able to do the user’s will like a human would." This vision implies a future where AI agents can intuitively understand complex intentions, manipulate computer interfaces, and autonomously complete multi-step tasks, mirroring human proficiency. Such a development would fundamentally transform human-computer interaction, enabling software to become more proactive and responsive to user needs without constant explicit instruction.

This ambitious outlook is increasingly supported by breakthroughs in AI. A prime example is Manus, a general-purpose AI agent that has garnered significant attention for its advanced capabilities. Manus recently demonstrated its prowess by scoring an impressive 86.5% on the Gaia benchmark, a rigorous test designed to evaluate AI agents on reasoning, multimodal handling, web browsing, and tool proficiency, nearing the human average of 92%. Its architecture, built on a multi-agent system, allows it to dynamically decompose complex tasks and leverage a wide array of tools to achieve objectives.

Y Combinator, under Tan's leadership, is actively fostering startups that align with this vision. The accelerator's Fall 2025 "Request for Startups" highlights key themes such as "Infrastructure for multi-agent systems" and "AI-native enterprise software," signaling a strategic focus on building the foundational layers for these advanced agents. This commitment extends to educational initiatives, including YC's inaugural AI Startup School launched in February 2025, which convenes leading figures like Elon Musk and Sam Altman to discuss the future of AI.

The progression towards human-like web agents promises profound implications for automation, operational efficiency, and the very nature of software. As AI agents become more sophisticated in understanding and executing user will, they are poised to make software increasingly self-supporting, reducing the need for manual intervention across various industries. This evolution signifies a pivotal moment in AI development, pushing the boundaries of what autonomous systems can achieve.