The Arrival of Manus: A Storm of Excitement

On March 8, 2025, Manus AI burst onto the scene, captivating the world with its promise of autonomous task execution. Described as a "general AI agent that turns thoughts into actions," Manus has everyone talking globally, with some referring to it as the second “DeepSeek moment.” It is designed to go beyond the question-and-answer paradigm of traditional large language models (LLMs) like ChatGPT, enabling users to delegate complex projects and workflows to an AI that can operate independently.

We have not yet gained beta access ourselves, but the examples on its website are impressive, as are the reviews on social media. Defining agents has been troublesome until now, creating an aura of mystique around the concept of AI agents. Manus changes that. Until agents arrived, AI mostly communicated—now it performs tasks. Manus acts as a smart helper that utilizes powerful language skills (from systems like LLMs or AI models) to understand user intentions, then determines how to execute them—perhaps by searching the web or filling out forms. Manus demonstrates that these agents are here to stay. Doubts about AI’s usefulness from a year ago? They’re fading fast as tools like this reveal what’s possible.

However, their arrival both simplifies and complicates the AI landscape. It signifies that AI isn’t just for tech wizards anymore—it’s for everyone, with LLMs serving as more than sophisticated information retrieval and generation tools. The possibilities with agents are limited only by our imagination, and the development of Manus suggests that these possibilities are closer than ever before.

The Flip Side: The Ease of Development in Building Agents

Given what Agents can do, we expect them to be in the headlines for many quarters to come. From an investment standpoint, the most important thing to note is that most Agents are extremely easy to build and will be replicated and improved upon quickly. While new features and implementations will continue to generate excitement - Manus will not be the last one - most Agent-creators will fade away equally quickly. Building, copying, and replicating Agents is far easier than building LLMs, and investors should be cautious if investing only in the builders for the features they may have touted.

Take Devin, an AI agent launched last year by a U.S. startup. It could write code and solve tech problems, wowing developers worldwide. Others built similar tools within days, tweaking Devin’s ideas to make them better or cheaper. Manus is facing the same fate. Already, posts on X suggest people are trying to mimic it, using free online tools. This speed shows how fast the agent race moves—great for users but tricky for companies.

Why is copying so simple? Building an agent doesn’t take much these days. Manus, for example, was created by a small team—maybe 20 or 30 people—in less than a year, with a budget under $10 million. Compare that to older AI projects like ChatGPT, which cost hundreds of millions and needed huge teams. Manus didn’t invent its brain from scratch; it used existing “smart engines” (like Anthropic’s Claude and Alibaba’s Qwen) and added clever programming to make them work together.

Building an AI Agent vs. Building an LLM: A Critical Comparison

To understand why AI agents are easier to replicate than foundational LLMs, it’s important to contrast the two:

LLMs require extensive pre-training on vast datasets, computational resources, and continuous refinement. AI agents, by contrast, rely on existing models and focus on integration, decision-making, and workflow automation—a much easier engineering problem.

This distinction is key to understanding why AI agents will proliferate rapidly. While OpenAI, DeepMind, and Google dominate foundational AI, agent development is far more accessible to startups, research labs, and independent developers.

So, while pioneers like Manus grab headlines, their edge might not last. Anyone with basic coding skills and a few thousand dollars can whip up a simple agent in weeks. Businesses will need more than cool tricks to succeed—they’ll need to find ways to keep users hooked or charge smart prices, which isn’t easy when copies flood the market.

Making People Pay USD20,000 per month Will Not Be Easy

Just as Manus burst onto the scene around March 8, 2025, whispers were already swirling about OpenAI’s next move. Rumors suggested the U.S. giant was preparing to roll out two new subscription packages—one at $2,000 per month for knowledge workers and another at a staggering $20,000 per month for “PhD-level research” capabilities. The timing couldn’t have been more striking, with Manus’s low-budget brilliance stealing the spotlight right as OpenAI aimed to set a new pricing bar. But charging such a hefty sum won’t be a walk in the park.

Subscription products thrive on stability and a constant edge over rivals. Customers expect a reliable service that keeps improving, with features that justify the cost month after month. For OpenAI’s $20,000 agent to succeed, it must deliver something extraordinary and stay ahead of the pack. Yet, unlike LLMs, which take years and billions to perfect, AI agents can spring up quickly. A great agent doesn’t need a tech superpower like the U.S. or China to birth it—it could come from a small team in India, a university in Europe, or even a lone coder in Brazil. This global accessibility spells trouble for high-priced offerings.

Even before the release of Manus, the GenAI landscape is already ablaze with demonstrations and promises of AI agent capabilities, many of which will undoubtedly materialize. However, the critical question, beyond all the features and capabilities, is different: which agents will possess the enduring uniqueness or technical complexity necessary to command premium pricing? From the investment viewpoint, it is less about features alone in the world of instant copyability. Even if certain features offer immense value to users, people won’t pay top dollar when similar options come at much lower costs.

Meanwhile, others are taking a different tack. Salesforce, for instance, launched an agent marketplace just days before Manus, betting on variety over a single, pricey product. The marketplace is designed to accelerate AI agent deployment by providing a library of templates and add-ons. This approach intensifies the competition for the creators, with the assumption that the platform provider may benefit by providing the infrastructure. 

But even other marketplaces are already bubbling up. Platforms like Hugging Face could pivot to host agent libraries, letting developers sell or share their creations. Smaller startups might launch niche hubs—one for education agents, another for healthcare—each vying for users. Even open-source efforts, like the buzz around OpenManus (a free spin-off inspired by Manus), hint at a future where agents are traded freely online. These marketplaces will keep sprouting, and hence, staying on top will be a constant scramble even for giants like Salesforce. 

The Real Risk from China: Pricing Power, Not Product Quality

China’s strength in AI isn’t just about technological breakthroughs—it’s about affordability. Manus AI, built on a budget of under $10 million, has initial pricing plans at least 90% lower than the rumored US$2000 per month plans of OpenAI. 

So far, despite the compute and other constraints, quality of GenAI products coming out of China appears at par, irrespective of the underlying compute constraints. Manus integrates leading models like Alibaba’s Qwen and Anthropic’s Claude, proving it can compete with Western AI on performance. The strategy is familiar—Xiaomi in smartphones, CATL in batteries, and Huawei in telecom. These companies didn’t just compete; they made high-end tech accessible at unbeatable prices, often forcing rivals to rethink their business models. Even when their output quality is lower than the best available from outside China, the pricing differences will ensure fierce competitive pressure for everyone globally.

Historically, the pricing pressure from China was mild in software- and service-related domains. GenAI era is giving rise to many new trends, and one of them has to be China’s rising role as a software provider.

Manus and Beyond: A Global Race with Big Questions

The launch of Manus AI is a landmark moment in the transition from querying-based AI to autonomous agentic intelligence. It proves that AI agents are not just possible—they are here. However, its arrival also raises profound questions about business sustainability, replicability, and long-term value.

More importantly, these innovations continue to come from the least expected corners. This isn’t just a China story—it’s global. Around the same time as Manus, Foxconn from Taiwan launched its own LLM, reportedly developed in just four weeks. As discussed above, the pace of innovation and the number of players in Agent developments are likely to be far higher than in the LLM developments.

As impressive as the capabilities, developing Agents is not a gold mine for everyone. Features like Manus’s—say, sorting resumes in hours—wow us now, but soon they’ll be common. 

‍