Manus AI and the Dawn of Agentic Computing B

Introduction: The Weekend Phenomenon

Something extraordinary happened in the AI world this weekend. A new agent called Manus exploded onto the scene, swelling its Discord membership to 138,000 in a matter of days. Invite codes reportedly sold for upwards of $7,000 on Chinese social media platforms. To understand the significance of this moment, we need to place it within the broader timeline of AI evolution—particularly following the DeepSeek moment that fundamentally altered perceptions about the pace and geography of AI innovation.

But make no mistake—Manus represents something more profound than a “China moment.” It heralds an “agentic moment”—our first genuine glimpse of what truly autonomous AI agents might accomplish. It’s forcing us to rethink what constitutes innovation in AI. Is it better models, or better integration? Is it theoretical capability, or practical implementation?

The most revolutionary products don’t just improve what exists—they redefine our expectations for what’s possible. What we’re witnessing may be comparable to the 2nd “ChatGPT moment” in AI: not the most powerful technology by raw specifications, but the most impactful through its integration and user experience.

Understanding Manus AI

At its core, Manus leverages a sophisticated multi-agent orchestration system. Unlike singular models, it employs an executor agent to interpret user commands, with specialised sub-agents handling planning, research, and execution tasks. The system integrates Anthropic’s Claude Sonnet (v3.5/3.7) and Alibaba’s Qwen models within a unified architecture.

What truly distinguishes Manus is its cloud-based asynchronous processing capability, allowing it to continue operating even when users disconnect. This creates the experience of a “digital employee” that requires only high-level directives to execute complex tasks.

Founder Yi Chow Peak G describes Manus as “bridging the gap between conception and execution,” positioning it as “the next paradigm of human-machine collaboration.” The Manus team claims top ranking on the GAIA benchmark with 82% accuracy, reportedly outperforming competitors including OpenAI’s Deep Research (68%) in AI autonomy, problem-solving, tool usage, and web interaction.

The system’s capabilities span diverse domains:

• Financial analysis: Generating detailed stock assessments with live market data integration
• Website development: Building functional websites from scratch, including resolving hosting issues
• Research automation: Producing comprehensive reports with citations and data visualisations
• Content creation: Designing multimedia presentations and educational materials
• Social media management: Simultaneously operating 26 accounts with context-aware posting

The Expert Reactions That Fuelled the Fire

The initial wave of reactions to Manus was nothing short of breathless. The Rundown’s Rowan Chung declared, “I think China’s second DeepSeek moment is here… It’s like Deep Research plus Operator plus cloud computer combined and it’s really good.”

Chung tested Manus by tasking it with creating his biography and deploying a website based on this information. He reported that “the info was 100% accurate with info up to date as of today,” emphasising this wasn’t any sort of paid endorsement.

Dean Ball offered an even more provocative assessment: “It is wrong to call Manus a DeepSeek moment. DeepSeek was about replication of capabilities already publicly achieved by American firms. Manus is actually advancing the frontier. The most sophisticated computer-using AI now comes from a Chinese startup. Full stop.”

Perhaps the most fascinating reaction came from McKay Wrigley, whose perspective evolved dramatically through his extended usage. His Twitter thread begins with cautious interest: “Watch a 14-minute demo of me using Manus for the first time. It’s shockingly good.” Hours later, his tone shifted dramatically: “All right, I’m starting to freak out a little bit… This experience has shifted my worldview… Literally thought this was going to be vaporware and now I’m amidst an existential crisis.”

These testimonials created a perfect storm of interest—combining authority, transformation, and FOMO in equal measure. When experts visibly shift from scepticism to evangelism in real-time, others take notice.

The Model Revelation and Its Significance

As the dust began to settle, an important revelation emerged: Manus uses Claude 3.7 Sonnet as its base model. This wasn’t some secret underground Chinese lab creation—it was built atop an existing, publicly available model.

This revelation matters profoundly because it frames Manus’s innovation as one of integration rather than model development. Ada McLaughlin articulated this perfectly: “If your opinion of Manus changed after discovering it’s a newer Sonnet wrapper and not some trained-up potatoes underground Chinese lab leak, you’ve lost the plot. I don’t care if it’s a wrapper, it created value, it deserves my respect. Care about capabilities, not architecture.”

This insight exposes a critical truth about the current state of AI development: we’ve entered an era where implementation may matter more than raw model capability. A huge proportion of what we’ll interact with from here forward will be experience or data wrappers sitting atop underlying models. LLMs have met the limits of scaling laws. Increasing data and compute is not enough to get the most out of AI. The true innovation lies in how these components and traditional system engineering are orchestrated to create unprecedented user experiences.

Reframing the “Manus Moment”

The discourse around Manus initially centred on geopolitical implications—another “China moment” following DeepSeek. But this framing misses the deeper significance.

The “Manus moment” isn’t about China versus the US; it’s about agents versus chatbots. It represents the first mainstream expression of what a genuinely agentic experience feels like—what AI can accomplish when freed from the constraints of a synchronous chat window.

This parallels the original ChatGPT moment. While experts had long understood the theoretical capabilities of LLMs, seeing them work in a fluid, accessible interface fundamentally altered public perception. Similarly, while AI builders have discussed agents extensively, Manus provides the first widely accessible demonstration of their potential.

The psychological impact of witnessing autonomous capabilities shouldn’t be underestimated. Seeing an AI assistant continue working on your behalf after you’ve disconnected—executing complex multi-step tasks without continuous guidance—creates a profoundly different relationship with technology.

The Tool-Intelligence Paradigm

Manus highlights a critical insight in AI development that experts like Yann LeCun have emphasised: tool mastery often trumps raw model capability in practical applications. While Claude Sonnet provides competent reasoning, Manus’s breakthrough lies in its orchestration of 29+ tools across browsers, APIs, and coding environments.

This creates multiplicative value through strategic integration. When tools empower models to perceive, manipulate, and verify their environment, emergent capabilities appear that surpass what the base model could achieve in isolation.

The open ecosystem advantage becomes evident here. Manus’s reliance on existing models like Claude and Qwen, alongside accessible tools, demonstrates how API access to diverse systems can democratise high-performance AI without requiring proprietary model breakthroughs.

However, we must avoid the false dichotomy between intelligence and tool access. Like orangutans using sticks to extract termites, these systems remain reactive tool-users rather than intentional tool-creators. True intelligence will require more fundamental advances in hierarchical planning and physics-based reasoning—but Manus shows how far the tool-integration approach can take us in the interim.

Current and Future Implications

As one commentator noted: “The models are already AGI-grade, and the last steps are how nicely we scaffold perception, context, memory, and the for-loop.” This may overstate current capabilities, but it captures an essential truth: much of what we need already exists in today’s models—we’re primarily experiencing a creative deficit in implementation.

The coming wave of innovation will focus on what creates value: scaffolding perception, context, and memory. As another observer noted, “We’re currently experiencing a human creativity deficit in AI development. We’re not building wrappers fast enough. User experience, context management, memory integrations, tool use—these are your moats.”

This represents a significant shift in how we measure progress. Token generation speed and parameter count will matter less than how effectively systems can maintain coherence across complex workflows. Companies that master the integration of perception, memory, planning, and verification will outcompete those focused solely on model performance.

We can expect costs to plummet while capabilities expand: “The cost per token will drop by 90% next year. There is no possible experience with LLMs made non-viable with cost or capability that will remain so in 12 months.” This acceleration means you shouldn’t build for today’s capabilities but for next year’s anticipated improvements.

Balanced Assessment

Despite the breathless enthusiasm, a balanced assessment reveals both Manus’s strengths and limitations.

Strengths:

• Research capabilities that produce comprehensive reports with proper citations
• Background script execution that enables autonomous operation
• Structured planning approach that breaks complex tasks into manageable steps
• Asynchronous processing allowing continuous progress during disconnection

Limitations:

• Processing speed issues, with operations described as “slow”
• Context window limitations causing breakdowns during extended operations
• Occasional task looping when the system encounters ambiguity
• Reliability challenges, particularly during complex coding tasks

AI for Success provided one of the most balanced perspectives after three days of testing: “The research it does on the Internet and the reports it generates are incredible. Its ability to run scripts behind the scenes to execute tasks is impressive. The plans it creates to achieve tasks are well-structured… [But] it’s slow… It could use a longer context window… It broke in between due to context issues while working on coding tasks.”

Strategic Industry Implications

The Manus phenomenon holds significant implications for the AI industry’s strategic direction.

First, tool-rich environments now represent a competitive advantage potentially more valuable than model development. Companies that build comprehensive tool libraries and integration frameworks may outperform those focused exclusively on model training.

Second, standardised API protocols will become increasingly important. Model Context Protocol could enter significant white space here. As systems like Manus demonstrate the power of coordinated tool use, the ability to quickly integrate diverse capabilities will become critical.

Third, we’re witnessing a shift from model development to experience design (traditional system engineering). While core model research remains vital, the competitive frontier is increasingly about creating intuitive interfaces that unlock existing capabilities rather than developing new ones.

This has profound implications for talent allocation and investment. Firms may need to rebalance resources from pure AI research toward user experience design, tool integration, and workflow automation. The skills required to build successful AI products are diversifying beyond machine learning expertise to include product design, API architecture, and human-computer interaction.

Conclusion: Beyond the Manus Moment

I predict that within just a matter of months, what we’re calling an agent now with Manus will seem quaint, barely autonomous, unsophisticated in its planning, and a far cry from what we’ll be using instead. The pace of innovation in the agent space is accelerating exponentially, driven by the realisation that much of what we need already exists—we simply need better ways to orchestrate it.

This pattern has played out repeatedly in technology history. The first smartphones appear primitive compared to today’s devices, yet they fundamentally altered our relationship with computing. Similarly, Manus may be remembered not for what it is, but for what it revealed about our AI future: that the path to more capable systems lies as much in integration as innovation.

The companies that thrive in this new landscape will be those that focus on user experience and practical implementation, not just technical capabilities. They’ll create agents that fit seamlessly into human workflows, anticipating needs and executing complex tasks with minimal supervision.

We’ll remember the Manus moment as the point when AI shifted from conversation to action, from responding to thinking, from assisting to doing. It’s not the destination—it’s the first step on a journey toward truly autonomous digital assistance that will transform how we work, create, and solve problems.

In the words of Brian Romney: “We just moved from chat AI era to agent AI era.” The Manus moment marks this transition—not because of what Manus is today, but because of what it shows is now possible.