Manus is being called the most impressive AI tool ever tried, showcasing its capabilities as a sophisticated computer using AI, unlike its predecessors that were merely specialized chatbots.

To measure Manus' capabilities, the Gaia benchmark tests AI agents on reasoning, multimodal handling, web browsing, and tool proficiency, with humans scoring about 92% and OpenAI's deep research scoring around 74%. Manus achieved an impressive benchmark score of 86.5% on the Gaia scoring system, just a few points shy of the average human performance.

Manus operates as a multi-agent AI system, coordinating a team of sub-agents to complete various tasks, from travel planning to financial analysis, rather than relying on a single neural network.

Some critics dismiss Manus as merely a wrapper for existing foundational models and tools, but this perspective overlooks the reality that many successful AI products also fit this description.

Manus exemplifies these trade-offs well, as its multi-agent orchestration significantly reduces per-task costs to around $2, compared to competitors like OpenAI's Deep Research.

The first cohort of AI's killer apps has emerged, including ChatGPT and others, indicating a significant shift in the market, while many new companies will adopt an agent-first approach, evolving from basic prototypes to robust agents capable of performing complex tasks.

Despite its strong performance, Manus has sparked a broader discussion about the nature of AI startups and the role of application layer wrappers.

The age of AI agents is here, and Anthropic's Claude Computer Use is one of the first to enable AI to browse the web, click buttons, and type text independently.

At Google DeepMind, we've been applying AI across almost every branch of science, achieving breakthroughs in areas from mathematics to life sciences, including developing AlphaProof for solving Math Olympiad problems and AlphaFold for predicting molecular structures.