What tech is behind the Manus Ai that is driving the world crazy?

Everyone wanted to be on their waiting list

CodeAct is a lesson everyone should learn from Manus AI Agents

Here's why Manus adopted its principles to build their agents....

When Manus AI was released,

Everyone was shocked by the number of tasks the agent could achieve with such great precision.

When asked by an X user if Manus uses MCP, the Co-Founder replied that they don't,

But have adopted a few principles of CodeAct that helped them make better decisions in tools as well as in execution.

What separates CodeAct from existing agent Architectures like ReAct or even Reflexion?

📌 Let us understand by breaking down its architecture:

1. Observation: The agent interprets the user's request and assesses the task's current state.

2. Agent: The input is first taken by the agent, which builds the primary foundation of what the solution should look like.

3. Planning: It then determines the appropriate tools or actions needed for the next step.

• Action: The first action taken by the Agent is shared with the codeAct Sandbox.

4. CodeAct Sandbox Manus writes Python code to perform the necessary action, which runs in a Linux sandbox environment.

• It evaluates multiple scenarios, selects the best action, and generates Python code as the universal action format for LLM agents.

• It utilises the given set of tools, databases and memory to build a capable solution.

5. Outcome: It displayed the action taken by the agent to the user as per their query.

• The outcome is not only sent to the user, but its feedback is also sent back to the agent to observe.

• Based on the observations, Manus can debug, adjust its approach, and try again if needed

This cycle repeats until the task is completed

Depending on your use case, CodeAct can provide output through natural language answers or environment interactions.

Specifically,

CodeAct uses executable Python code for complex, multi-step operations,

Unlike other architectures limited to predefined functions or JSON, this approach simplifies debugging and refinement while keeping context length manageable.

(Note: The team hasn't detailed CodeAct's implementation but shared three key insights from applying its principles, available in the referenced comments.)

Author

AiUTOMATING PEOPLE, ABN ASIA was founded by people with deep roots in academia, with work experience in the US, Holland, Hungary, Japan, South Korea, Singapore, and Vietnam. ABN Asia is where academia and technology meet opportunity. With our cutting-edge solutions and competent software development services, we're helping businesses level up and take on the global scene. Our commitment: Faster. Better. More reliable. In most cases: Cheaper as well.

Feel free to reach out to us whenever you require IT services, digital consulting, off-the-shelf software solutions, or if you'd like to send us requests for proposals (RFPs). You can contact us at [email protected]. We're ready to assist you with all your technology needs.

© ABN ASIA