Manus: the frontier of Agentic AI

The Dawn of Agency: Manus AI’s Vision for the Future of General AI

Mr. Tao Zhang, Co-founder and CPO of Manus AI, addressed Web Summit Lisbon 2025, defining a true AI agent. He clarified that an agent possesses “agency”—the inherent ability to think, perform actions, and learn autonomously—a critical distinction from simple chatbots that merely follow system prompts.

Manus AI pioneers the “general agent” concept, designed to tackle a broad spectrum of tasks, moving beyond specialized vertical applications. Its unique agentic architecture operates without predefined workflows, employing a “gigantic agentic loop” powered by foundational models. This system dynamically plans, predicts “atomic capabilities” (like editing files, running commands, or browsing), and executes them within a fully functional virtual machine, providing the AI its own complete computational environment.

After each action, Manus gathers observations, feeding them back into its agentic loop. This continuous feedback enables the system to learn by doing, intelligently determining subsequent steps and solving diverse problems effectively. Manus AI has rapidly become a leader, topping Scale AI’s RFI benchmark for real-world problem-solving, with Mr. Zhang projecting significant growth in task resolution from 2.5% to 10-20% next year.

Since its launch in March, Manus AI has achieved a global user base and $100 million in annualized revenue within seven months. Mr. Zhang outlined three key frontiers for future agentic AI development: expanding atomic capabilities, enhancing foundational infrastructure, and fostering proactivity.

The first frontier involves integrating more atomic tools, creating a “network effect” where each new capability unlocks numerous scenarios (e.g., the “view image” tool enabling self-correction of website layouts). The second focuses on building robust infrastructure for complex, long-running tasks, decomposing them into independent, concurrently running subtasks to enhance reliability and speed, facilitating “wide research.”

The third frontier is proactivity, where agents operate continuously, anticipating and completing tasks before explicit requests. Mr. Zhang stressed a crucial mindset shift for adoption: focusing on “enhancing human with AI” rather than “replacing human with AI.” Key barriers include user apprehension about sharing personal context and developing AI-specific tools. He advised consistently assigning tasks to Manus to train users in “co-piloting with AI,” a skill he deems the “new literacy” for scaling human intelligence. Manus AI’s mission is to extend human reach and amplify human intelligence.