The sound of rain hitting the cobblestones of Lisbon echoed against the massive roof of the MEO Arena.
Unusually for November, thick clouds covered the sky, bringing the first rain to Web Summit in years.
I’m CityNomix. While I usually wrestle with numbers in the digital marketing world, right now I’m immersing myself in the forefront of technology here in Lisbon, Portugal.
The excitement from yesterday hasn’t faded yet. I was deeply impressed by Qualcomm’s vision of “Agent-Centric AI” (read more here) and nodded in agreement with Microsoft’s discussion on the challenges of AI adoption (read more here). After recharging with some exquisite Portuguese cuisine at Páteo Restaurante (restaurant review here), I was ready for day two.
And so, the second day began. Despite the unfortunate weather, the enthusiasm inside the venue was hot enough to evaporate the rain.
I spent the morning watching several startup pitches, but honestly, none delivered a knockout blow.
When lunchtime arrived, I headed to the food truck area only to find the indoor stalls packed with long queues due to the rain.
“Oh no,” I muttered, heading to a relatively less crowded outdoor food truck. Balancing an umbrella, I managed to grab a simple cheeseburger.
Eating a burger in the rain (read about day 2 lunch here) wasn’t exactly elegant, but it’s part of the authentic travel experience.
After a quick meal, I rushed to the “AI Summit” stage.
I had a hunch that the upcoming session would be the highlight of the day, if not the entire Web Summit.
The AI Summit Frenzy: Manus Discusses “The Frontier of Agentic AI”
When I arrived, the venue was already packed to the rafters.
I was prepared to stand, but luck was on my side as I snagged a seat just as the previous session ended.
On the screen, the words “The Frontier of Agentic AI” loomed large.
Taking the stage was Tao Zhang, Co-Founder & CPO of the buzzy Singaporean startup, “Manus.”
Amidst the current AI boom, many companies claim to offer “AI Agents.”
However, Zhang’s presentation was distinct from mere feature showcases or buzzword bingo.
From the outset, he posed a fundamental question to the audience.
What is an AI Agent? Breaking Away from Chatbots
“When we talk about ‘AI Agents,’ what comes to mind?”
Zhang sharply pointed out the misconceptions surrounding many “agents” flooding the market today.
“Chatbot + System Prompt ≠ Agent”
The moment this slide appeared, the atmosphere in the room shifted.
Simply feeding a chatbot a prompt like “You are an excellent assistant” does not make it an agent.
It remains, fundamentally, a tool waiting for instructions.
So, what is a True Agent?
He defined the answer with a single word:
Agency.
The ability to think, plan, and act autonomously without waiting for commands to proactively solve problems.
This is the destination manus ai is aiming for, and it is the form of AI we seek as a true partner.
Defining the General Agent
He further touched upon the concept of a “General Agent.”
Not an automation tool specialized for specific tasks, but a system that is adaptable like a human, versatile, and capable of solving open-ended problems while continuously learning.
That is the vision of Manus.
This definition goes far beyond what we typically think of as “AI for business efficiency.”
It’s not just about auto-replying to emails or scheduling appointments.
It’s about “intelligence” that seeks and executes solutions to unknown challenges on its own.
How Manus Works: Think, Act, Learn
So, how exactly does Manus function?
Zhang positioned Manus not just as an application but as a “problem-solving engine,” explaining its process in three steps.
-
- Think: Breaks down complex goals into executable phases.
-
- Act: Dynamically selects and executes the right tools, such as writing code, browsing the web, or editing files.
-
- Learn: Evaluates results, iterates, and optimizes to improve accuracy.
By cycling through this process at high speed, Manus achieves phenomenal performance.
In fact, according to Scale AI data, Manus records the highest fully autonomous task completion rate compared to major AI models like Gemini 2.5 Pro, GPT-5, and Sonnet 4.5.
However, as the graph shows, the full automation rate is still below 4%.
This indicates that the entire industry is still in the “early stages,” but Manus is clearly a head above the rest.
This fact alone explains why attention on manus ai is skyrocketing.
The Three Frontiers of Agentic AI
Zhang outlined three boundaries, or frontiers, in the evolution of Agentic AI.
Frontier 1: Atomic Capabilities
This is the stage of expanding what an agent can do.
Content creation, browser automation, coding.
In the future, it will include full control over desktop software.
Frontier 2: Foundation Architecture
Even with increased capabilities, they are meaningless without a strong foundation.
Robust infrastructure capable of decomposing complex tasks and executing parallelism is essential.
Here, he introduced data comparing “Wide Research” capabilities.
For research tasks that take humans days or cause standard chatbots to fail after a few items due to context saturation, Manus processes hundreds of items in minutes via parallel processing, maintaining uniform quality.
“Most AI tools fail at scale. Manus doesn’t.”
This statement was backed by technical confidence and proof.
Frontier 3: Proactive Agency
And finally, the ultimate frontier.
Acting “Even Before You Ask.”
Understanding user intent and preparing in advance.
This is the completed form of Agentic AI and the future Manus is striving for.
Shifting Mindset: Enhancing Humans
After the technical deep dive, Zhang spoke about our own “mindset.”
This was the most striking part of the session for me.
“Replace human? Or enhance human?”
Faced with AI evolution, many fear their jobs will be taken away (Replace).
However, Zhang strongly urged leaders to adopt an “Enhance Human” perspective.
“How can we make top performers 10x better?”
“What new possibilities will this create?”
Instead of calculating replacement rates and risks, we should focus on benefits and potential.
Collaborating with AI agents is the “new literacy of our era,” breaking down cognitive and creative barriers, just as cars overcame physical distance and computers overcame information walls.
Deep Dive: 8 Keywords to Understand Manus AI
Now, from CityNomix’s perspective, let’s thoroughly explore the key points about Manus AI that you, the readers, are likely curious about.
Manus AI Country of Origin
Manus is a startup based in Singapore.
Taking the main stage at an international tech conference like Web Summit signals that its technological prowess is gaining global attention.
Born in the innovation hub of Singapore, this Asian startup seems to take a different approach from Silicon Valley giants, prioritizing practicality and robustness.
Manus AI Risks
With autonomous AI, everyone worries about Manus AI risks: “Will it go rogue?” “Will it perform unauthorized actions?”
In his presentation, Tao Zhang acknowledged that “Barriers to Cross” like sharing context and exposing tools require “courage.”
However, Manus incorporates logic for appropriate tool selection in the “Act” phase and employs an architecture that performs parallel processing safely under user supervision.
While risks aren’t zero, the design philosophy focuses on making them controllable while maximizing the benefit of human enhancement.
Manus AI Safety
Regarding Manus AI safety, the “Foundation Architecture” is key to meeting corporate security standards.
Manus appears to target enterprise use, aiming for high security standards in data handling and system access management.
Features like “Wide Research,” which involve accessing external sites, are likely executed in secure sandbox environments to prevent systemic threats.
Manus AI Free
Many readers are curious about a Manus AI free plan.
Currently, Manus is positioned as a high-functionality tool for business and professional use.
There is no massive promotion of a completely free plan.
Given the costs associated with advanced parallel processing and API integrations, it will likely follow a paid subscription or usage-based model.
However, keep an eye out for beta access or waiting lists on their official site.
Manus AI Image Generation
The presentation slide on “Frontier 1: Atomic Capabilities” listed “Content creation (images, videos…)”.
This means Manus AI image generation is included as part of its capabilities.
Rather than being a specialized tool like Midjourney, it will likely call upon image generation AIs as part of a larger workflow—for example, generating an image to insert into a blog post it is writing.
Manus AI Japanese Support
Although the Web Summit presentation was in English, the probability of Manus AI Japanese support is extremely high since it is based on modern LLMs.
If the underlying model supports multiple languages, users can instruct Manus in Japanese during the “Think” and “Act” phases and receive results in Japanese.
Given Japan’s active AI market, we can expect Japanese UI and support upon or shortly after official release.
How to Use Manus AI
The basics of how to use Manus AI go beyond chatting.
It starts with giving a “Goal.”
For instance, you might instruct: “Research pricing for 100 competitors and summarize it in a spreadsheet.”
Manus then:
1. Think: Decomposes the task into search, extraction, and aggregation.
2. Act: Launches a browser, visits sites, and retrieves data.
3. Learn: Corrects errors and compiles the report.
The user monitors the process and receives the final output. This is the new way of interacting with agents.
Manus AI App
While currently deployed as a web-based platform, future “Atomic Capabilities” include desktop control.
We can envision a Manus AI app for desktops to automate PC operations.
Imagine instructing Manus from a smartphone app while out, and having it complete tasks on your cloud or desktop workspace.
Conclusion: Building the Future with Manus
“Join Us in Building the Future.”
Tao Zhang concluded by calling out to all builders in the venue.
Listening to the Manus session at Web Summit, I felt something more than just tool evolution.
It’s the arrival of an era where humans are freed from tedious tasks to focus on genuine “creation” and “thinking.”
This excitement witnessed in rainy Lisbon might be spoken of years later as “the turning point.”
Manus’s mission is “Extending Human Reach.”
We, too, should not fear AI but rather embrace the challenge of discovering how far we can expand our own potential with this new partner.
I hope this report provides insights into AI agents and hints for your next steps.
This was CityNomix.
