Manus AI, Zoom's AI Companion, and OpenAI's New Developer Tools

Imagine an AI agent that not only responds to your queries but independently makes decisions and completes tasks without any human input. Welcome to The AI Agent Daily Brief, your go-to for the latest AI updates. Today is Monday, March 17th, 2025. Here’s what you need to know about Manus AI, China’s groundbreaking fully autonomous AI agent. Let’s dive in.

China's Manus AI, developed by the company Monica, has taken the AI world by storm with its launch on March 6th, 2025. Unlike reactive models like ChatGPT that rely on prompts, Manus AI is designed to operate autonomously, executing tasks and delivering results with minimal human oversight. This marks a significant shift in AI development, moving from reactive systems to truly autonomous agents.

The name 'Manus' comes from the Latin phrase 'Mens et Manus,' which translates to 'Mind and Hand.' This aptly describes its dual capabilities: thinking and acting. Manus leverages large language models, such as Claude 3.5 Sonnet from Anthropic and Qwen from Alibaba, for processing complex information and decision-making, while integrating these with traditional automation tools for execution. What makes Manus AI stand out is its hybrid neuro-symbolic approach.

It combines the creativity of generative AI with the dependability of deterministic scripts. For example, while a language model might draft Python code, Manus ensures that it executes this code in a controlled environment, validating and adjusting as needed. This allows Manus to handle complex tasks like deploying web applications or automating interactions across platforms. Manus operates through a structured agent loop that mimics human decision-making.

When assigned a task, it analyzes the request, identifies objectives, and selects the necessary tools. It then executes commands in a secure Linux sandbox, ensuring safe operation. After each action, Manus evaluates outcomes and refines its approach until it achieves the desired results. One of the defining features of Manus AI is its multi-agent architecture. A central "executor" agent manages various specialized sub-agents, each handling specific tasks like web browsing or data analysis.

This structure allows Manus to tackle multi-step problems without additional human input, working asynchronously in the cloud and delivering results once tasks are complete. In terms of performance, Manus AI has excelled in the GAIA Benchmark, developed by Meta AI, Hugging Face, and AutoGPT. This benchmark tests an AI's ability to reason, process multi-modal data, and execute real-world tasks.

Manus's performance has outstripped even giants like OpenAI's GPT-4 and Google's models, solidifying its place as a leader in general AI agents. Manus AI's practical capabilities were showcased in several use cases during its launch. For example, it autonomously handled a hiring process by analyzing resumes and market trends, generating a comprehensive hiring report without human intervention.

It also created personalized travel itineraries and even developed a website, all while considering intricate details like user preferences and external factors. However, despite its impressive capabilities, Manus AI faces challenges. Users have reported instances where the system enters ineffective loops, requiring human resets. Its web automation features raise ethical concerns about potential misuse, such as data scraping or platform manipulation.

Moreover, transparency issues remain, as independent verification of its full capabilities is limited, making trust-building a challenge. In summary, Manus AI represents a new frontier in AI: autonomous agents capable of executing tasks across industries independently. As these systems advance, they could redefine industries, reshape labor markets, and challenge our understanding of work itself.

Manus is just the beginning, heralding an era where AI doesn't just assist but independently acts and learns.

OpenAI's latest developer tools are making waves in the AI community. Earlier this week, they released two new solutions designed to help developers build AI agents that go beyond just answering questions and can actually take actions, like approving refunds or buying plane tickets. Imagine an AI that doesn't just give you information, but also makes things happen—it's a game-changer.

In a live stream on Tuesday, March 11th, OpenAI introduced the Agents software development kit, or SDK, along with the Responses application programming interface, or API. These tools provide a powerful framework for developers looking to create applications that can access a wide range of capabilities. Think of the SDK as the spell book and the API as the magic wand. The SDK gives developers the language needed to call upon these new powers, while the API channels those powers to applications.

During the demonstration, OpenAI employees showed how easy it is to use these tools by creating an AI stylist agent. This agent could search the internet and access a private database to recommend stores with products matching personal style preferences. They even created a customer support agent that can search through order databases and submit refund requests. To top it off, they made a third agent that seamlessly switched between the stylist and customer support agents based on context.

Olivier Godement, who leads OpenAI's API business, mentioned that within just 24 hours of the SDK and API announcement, several companies had already launched new products. This shows how quickly businesses are adopting these tools to build more advanced AI agents. OpenAI's chief commercial officer, Giancarlo Lionetti, shared examples of customers developing billing agents, financial analyst agents, and even application agents for automating user enrollment processes.

Box, a cloud-based data storage platform, was one of the first companies to leverage these new tools. Within 48 hours, Box's team built an agent that connects proprietary data on their platform with OpenAI's models to accomplish various tasks. Picture an agent that automates customer service tasks like approving or denying refunds by accessing specific data on Box.

Box is even redesigning its internal search functions to be more agent-friendly, ensuring agents can access the most relevant information. OpenAI's Godement expects businesses to become more ambitious with their agentic use cases over the coming months. Imagine a future where your personal agent, with access to your identification and credit card information, can make transactions just by talking to a merchant's agent.

Instead of spending time hunting down a new sweater online, your personal agent could handle everything for you. However, there's a word of caution: don't overload agents with too much information. Kus advises creating specialized agents for each step of a workflow to ensure they're effective and focused.

Kagent is the new kid on the block in the open source arena, and it’s here to make a splash in the Kubernetes world. Solo.io has just introduced this open source framework specifically designed to help developers and platform engineers build and run AI agents that can supercharge Kubernetes workflows. If you’re in the DevOps space, this could be a game-changer for automating tasks like configuration, troubleshooting, and even network security.

Now, imagine you’re a platform engineer trying to keep everything running smoothly amidst a storm of requests. Kagent steps in as your digital assistant, ready to handle the grunt work. It integrates seamlessly with other cloud native tools, thanks to its architecture built on the Model Context Protocol. This protocol, introduced by Anthropic, aims to standardize how AI models talk to APIs, making Kagent a versatile addition to your toolkit. Why does this matter?

Because Kagent is built on Microsoft’s open source AutoGen framework and carries an Apache 2.0 license, meaning it’s free to use and contribute to. Lin Sun from Solo.io explained that the project originated as a way to solve internal challenges, particularly during a crisis when an insurance company needed urgent troubleshooting help. This framework was born out of necessity, and now it’s poised to help others facing similar challenges. The real beauty of Kagent lies in its initial offerings.

It comes with tools for Argo, Helm, Istio, and Kubernetes, along with observability tools like Grafana and Prometheus. But that's not all—it also offers a cloud native expert knowledge base that can expand with any MCP-compatible tool server. The framework is designed with three layers: tools, agents, and a declarative API and controller. This setup allows users to build and run agents through a user interface, command line interface, or declarative configuration.

Lin Sun has big dreams for Kagent, hoping it’ll inspire the community to build on what Solo.io has started. They’ve seeded the project with some sample agents and tools, integrated with Kubernetes, and now they’re inviting the community to enhance it further. Sun envisions a future where Kagent has agents for every critical cloud native project, making it easier for new users to navigate the landscape with project-specific agents at their side.

The plan is to donate Kagent to the Cloud Native Computing Foundation, following in the footsteps of Solo.io’s previous donation of Gloo Gateway. This means Kagent could soon become a staple in the open source community, further enriched by the contributions of developers around the world. What’s on the wish list for the future? Sun mentioned the desire for more tracing capabilities, integration with OpenTelemetry, and enhanced metrics for Kagent.

They’re also looking to add multi-agent support and expand support for different large language models beyond OpenAI. If you’re interested in contributing, check out the CNCF Slack’s #kagent channel or stop by Solo.io’s booth at KubeCon + CloudNativeCon Europe 2025.

Zoom is stepping up its game by adding even more artificial intelligence agents across its platform. In a recent press release, Zoom announced that its AI Companion has evolved from being just a personal assistant to a truly agentic tool. This means it's taking a big leap forward in how AI can boost productivity and collaboration at work.

Imagine having an AI that not only helps you schedule meetings but also finds a time that works for everyone, generates clips, and even assists with advanced document creation. It's like having a digital assistant who never clocks out. Zoom's AI Companion is now offering virtual agents for customer self-service within its Business Services. In the coming months, it'll also add features to create and deploy customizable virtual agents.

These agents will aid sales teams with automated insights, personalized outreach, and enhanced prospecting. But that's not all—soon, AI Companion will be able to work with third-party and custom agents to streamline sales requests for proposals, information technology service requests, and human resources service requests.

Smita Hashim, Zoom's Chief Product Officer, highlighted the value these AI agents bring to customers, helping them connect, collaborate, and get more done—all within the trusted Zoom platform. The focus on AI has allowed Zoom to expand its offerings, enter new markets like the contact center space, and stand out from competitors. Embedding AI across Zoom's Workplace and Business Services is a top priority, driving productivity and engagement for their users.

This move by Zoom comes amid a broader trend in the tech industry. Just this month, Meta announced plans to expand its AI offerings for business users, and Amazon Web Services launched a new group focused on agentic AI. It's clear that AI agents are becoming a crucial part of the digital transformation landscape, offering businesses new ways to enhance their operations.

That’s it for today’s The AI Agent Daily Brief. Manus AI is leading the charge with its autonomous capabilities, setting a new standard in AI development. Thanks for tuning in—subscribe to stay updated. This is Michelle, signing off. Until next time.