First Look: Manus - The new AI agent

If you've been following AI developments, you've probably noticed a recurring pattern: tools that promise to build full applications but deliver disappointingly limited results.

But what if I told you there's finally an AI agent that can actually create functional apps, perform complex research, and interact with the web in ways that feel genuinely useful?

Let's explore Manus.im, an AI agent that's changing what's possible in the no-code/AI-assisted development space.

What is Manus?

Manus is a new AI agent built by a team of Chinese researchers using Claude (not the latest model, which makes this even more impressive). It features a sandbox environment that gives it access to:

• A browser it can control
• A full development environment
• The ability to deploy applications
• Web scraping capabilities

Unlike many AI assistants that just generate code or explain concepts, Manus can actually execute tasks in a virtual environment, allowing it to build and test applications in real-time while you watch.

What can it do? Let's see it in action

During our workshop, we put Manus through several tests to see what it could handle:

1. Building a collaborative whiteboard app

When asked to "build a collaborative whiteboard application," Manus immediately:

• Created a project directory
• Researched real-time collaboration technologies
• Built a tech stack with React, Fabric.js, Node.js and Socket.io
• Implemented the entire application
• Deployed a functional version

The most surprising part? It actually worked. Multiple participants were able to simultaneously draw on the shared whiteboard through a link Manus provided. While not the prettiest interface, it was completely functional.

2. Scraping financial data

We asked Manus to retrieve specific financial information from CF Data for a particular security (NMFC). It:

• Navigated to the website
• Searched for the ticker symbol
• Located the premium/discount information (-11%)
• Successfully extracted the data

What's particularly impressive is that it handled this without human intervention, even navigating through potential CAPTCHA challenges.

3. Financial market analysis

In a previous session, I had asked Manus to "find patterns in the news that preceded the most recent big stock falls and gains over the last six to eight months."

Manus created a comprehensive report that:

• Defined what qualified as significant market movements
• Collected historical market data
• Researched news events surrounding each movement
• Compiled findings showing patterns (like policy uncertainty preceding market drops)
• Generated data visualization and analysis scripts

The depth of analysis produced in minutes would have taken hours of manual research.

4. Building a GitHub guide generator

I asked Manus to create an app that would take a GitHub URL and generate a simple guide for non-technical users. It:

• Built the application
• Connected to my GitHub (when given an access token)
• Pushed the finished code to my GitHub repository
• Generated documentation

While the guide wasn't quite as non-technical as I'd hoped, the ability to deploy directly to GitHub was impressive.

5. Scraping and populating Google Sheets

As a final test, I asked Manus to scrape the latest posts from Ben's Bytes newsletter and populate a Google Sheet with the titles and URLs.

It navigated to the site, extracted the information, and then—after being given editor access—populated a Google Sheet with the data, typing it in cell by cell.

What makes Manus different?

1. It can interact with the web: Unlike many AI tools that can only generate content, Manus can actually navigate websites, fill forms, and extract information.
2. It builds end-to-end solutions: Rather than just generating code snippets, it builds complete, functioning applications.
3. It can deploy what it builds: Through its sandbox environment, it can deploy applications and provide access links.
4. It handles background tasks: You can set Manus on a task and check back later, much like having a developer working for you.
5. It integrates with existing tools: From GitHub to Google Sheets, it can work with tools you already use.

The practical impact

For non-developers, this is a game-changer. Tasks that previously required hiring a developer or learning to code are now accessible through simple prompts.

For example, a non-technical founder could:

• Build an MVP to test an idea
• Create custom tools to analyze data
• Automate workflows between different platforms
• Generate reports from scattered data sources

Even for those with technical skills, Manus acts as an accelerator, handling the implementation details while you focus on the bigger picture.

Limitations and considerations

While impressive, Manus isn't perfect:

• It occasionally struggles with complex projects
• The context length and limits are quite restrictive in its beta roll-out
• Some tasks take longer than expected
• There are security considerations when providing access to your accounts

Getting started

Currently, Manus is in limited release with an invite system. You can join the waitlist at manus.im.

The bottom Lline

Manus represents a significant step forward in AI agents that can actually build useful things. It bridges the gap between "I have an idea" and "I have a working product" in ways that were simply not possible until recently.

What makes this even more exciting is that this is just the beginning. As these tools continue to evolve and integrate with more services, the barrier between having an idea and implementing it will continue to shrink.

The era of "vibe coding"—where you describe what you want and AI builds it—is no longer just a meme. It's becoming reality, and Manus is leading the way.