Usable AI agents are finally here, with platforms like OpenAI's Deep Research and Google's tools leading the way, and now Manus has joined the competition as a brand new general-purpose AI agent that has taken the world by storm.

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.

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.

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 offers a glimpse into the future of AI, potentially allowing users to run a desktop operating system directly on their computers, providing more control than browser-based solutions.

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

Manus can accomplish a wide range of real-world tasks, excelling in creating travel itineraries, detailed financial analyses, and educational content, while also assisting with database compilation and high-quality presentations.

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.

One innovative feature of Manus is its ability to expose the file system, enabling users to see exactly what the agents are doing, unlike ChatGPT, which requires reprompting and lacks transparency.