Manus AI - Autonomous AI Agent

Hello AI enthusiasts,

Have you ever thought about this?

Most content, whether written by AI or humans, is currently aimed at humans, not LLMs.

Very soon, this focus on human attention will shift to LLM attention.

For example, 99% of libraries still have docs that render as pretty .html static pages, assuming a human will navigate them. In 2025, those docs should be a single your_project.md text file designed to fit into an LLM’s context window.

Rinse and repeat for everything else.

In this episode: Manus AI, OpenAI AI agents, GDM robotics models, Ray2, 20k agents(?)

Manus AI - China’s great AI Agent export

Manus AI has shaken up the AI world. Another Chinese company delivering a leading frontier AI application.
It’s an AI agent that independently researches for you, conducts competitive analysis, and creates websites instantly for reports and dashboards.

I first came across it through this post—Manus AI at scale:

But even more impressive is how powerful and capable Manus AI is.
Vibe coding—where you iteratively build software with AI agents, guiding them step-by-step—becomes a thing of the past with Manus AI.
It doesn’t need multiple iterations; Manus AI creates complete solutions from a single instruction.

For example, with this prompt: "Make a three.js endless runner game with a Schwarzman Scholars theme," Manus AI built a fully functioning game in one shot.

Getting access is not that easy as of now (they open up gradually). However, join the waitlist here: https://manus.im 

There are a couple of copycats out there, such as Openmanus, but nothing comes close to its performance yet.

Interestingly, as I mentioned last week about Alibaba's latest language models QwQ, Manus is collaborating with them in a Chinese partnership.

OpenAI’s new tools for building AI Agents

OpenAI’s new releases are all about building AI agents, simply. They have released:

• the Responses API, blending Chat Completions and Assistants APIs. It includes built-in OpenAI tools auto-executing calls and adding results to context.

• New Tools: web search, file search and Agend SDK. Agent SDK is an open-source framework for easy agent design/scaling, incl. observability for performance tracking.

Agent SDK is the successor version of swarm AI. As I currently have a project that is built on top of swarm, I will upgrade it, and share the learnings with you.

Find the quick start here.

Google DeepMind’s Gemini Robotics Models

On the 12th of March, Google DeepMind (GDM) enhanced Gemini 2.0 with robot capabilities. It includes advanced reasoning, interactivity, dexterity, and generalization.

With that, robots can perform e.g. origami, and with the support of a wide range of robotic platforms, including bi-arm systems like ALOHA 2 and Franka, as well as humanoid robots like Apptronik’s Apollo, it is interoperable on scale.

They have opened their waitlist.

Robots (not only humanoids) is a fiercely competitive market with US and Chinese companies on the top. I am inclined to say China with Unitree might be number 1 in terms of performing complex tasks, while actually selling robots.

Introducing Luma - Ray2

OpenAI plans to introduce $20k/month PhD-level AI agents

It would likely be capable of performing highly specialized, expert-level tasks in academic research and software development.

It could autonomously conduct advanced analyses, write and debug complex code, synthesize research, and solve difficult problems across various domains—all with the reasoning and precision of a PhD-level human expert.

I hope you enjoyed it.

Happy weekend!
Martin 🙇

• I recommend:

  • Beehiiv if you write newsletters.

  • Superhuman if you write a lot of emails.

  • Cursor if you code a lot.

  • Bolt.new for full-stack development.

  • My book - Generative AI: Navigating the Course to AGI.

  • Follow me on X.com.

• AI for your org: We build custom AI solutions for half the market price and time (building with AI Agents). Contact us to know more.

You might like our last episodes: