Join our daily and weekly newsletter for the latest updates and exclusive content on industry-leading AI coverage. learn more
The universal AI Agent landscape suddenly became more crowded and ambitious.
This week, a Palo Alto-based startup Genspark Released the so-called Super AgentThis is a fast-moving autonomous system designed to handle practical tasks in various fields – including some people who cause eyebrows, such as calling restaurants using realistic synthetic sounds.
This launch adds fuel to the important new front in the AI race: Who will build the first universal agent that is reliable, flexible and truly useful? Perhaps more urgent, what does this mean for the business?
Genspark’s super agent launches startups created only in a different Chinese language, Manus,,,,, Gaining attention for its ability To coordinate tools and data sources to complete asynchronous cloud tasks such as travel booking, recovery filtering, and inventory analysis – all without the typical hand-knots for most current agents.
Genspark now claims to go further. According to co-founder Eric Jing, Super Agent is built on three pillars: a concert of nine different LLMs, over 80 tools and over 10 proprietary datasets, all of which work together in a coordinated process. It goes far beyond traditional chatbots, handling complex workflows and returning fully executed results.
exist DemoGenspark’s agent plans a full San Diego trip, calculates the walking distance between attractions, maps public transport options, and then uses voice calls to an agent to book restaurants, including handling food allergies and seating preferences. Another demonstration shows the agency creating a cooking video reel by generating recipe steps, video scenes, and audio overlays. In one third, it wrote and produced an animated episode in South Park style, recreating the recent Signal Gate political scandal involving sharing war plans with political journalists.
These may sound consumer-centric, but they show where the technology is heading – moving towards multi-modal multi-step task automation, which blurs the line between creative generation and execution.
“Solving these real-world problems is much more difficult than we think,” Jing said in the video.
A fascinating feature: The super agent clearly expresses its thinking process and tracks the reasons through each step, which tools it calls, and why. Watching this logic work in real time makes the system feel less like a black box, but more like a collaborative partner. It can also inspire enterprise developers to build similar Traceable reasoning paths Enter your own AI system to make your applications more transparent and trustworthy.
Super agents are also easy to try. The interface starts smoothly in the browser without technical settings. GensPark allows users to start testing without personal credentials. By contrast, MANUS still requires applicants to be on the waitlist and disclose social accounts and other private information, thus increasing experimental friction.
We first published an article about Genspark in November Cloud-driven financial reporting. It has At least $160 million was raised in two roundsand is supported by us and Singapore investors.
Watch the latest news A video discussion between AI agent developer Sam Witteveen and me here To gain insight into the comparison of Genspark’s approach to other proxy frameworks and why it is crucial to enterprise AI teams.
How to delete it in Genspark?
Genspark’s approach stands out because it has led to a long-standing AI engineering challenge: large-scale tool orchestration.
Most current agents crash when having both yarn, not just a few external APIs or tools. Genspark’s super agent seems to be able to better manage this by using model routing and retrieval-based selection to dynamically select tools and sub-models based on tasks.
This strategy echoes emerging research Cotools, a new framework from Soochow University in China This enhances the extensive and evolving tool set of LLM. Unlike older methods that rely on rapid engineering or strict fine-tuning, Cotools keeps the basic model “freezing” while training smaller components to judge, retrieve and call tools to work.
Another pusher is the Model Context Protocol (MCP)little known but Standards are increasingly adopted This allows the agent to carry more richer tools and memory environments across steps. Combined with Genspark’s proprietary dataset, MCP may be one of the reasons its agents emerge More “coordinated” than alternatives.
How does this compare to Manus?
Genspark is not the first startup to promote a general agent. Manuslaunched last month by China-based Monica, waving waves using its multi-agent system that automatically runs tools like a web browser, code editor or spreadsheet engine to complete multi-step tasks.
Manus effectively integrates open source parts, including LLMs such as Anthropic’s Claude, including web tools and LLM, which is surprising. Although no proprietary model stack is built, it still performs well on the Gaia benchmark, a synthetic test designed to evaluate the agency’s actual task automation.
However, Genspark claims to have surpassed Manus, scoring 87.8% on Manus’ reported Gaia and using buildings that include proprietary components and a wider tool coverage.
Big tech player: Still playing safely?
At the same time, the largest AI company in the United States has been cautious.
MicrosoftThe main AI agent, Copilot Studio, Focus on fine-tuned vertical proxy that is closely aligned with enterprise applications such as Excel and Outlook. Openai‘ The Agent SDK provides building blocks, but lacks the full functionality to ship itself. General agent. AMAZonIt was recently announced that Nova Act took a developer’s first approach, providing atomic browser-based actions through the SDK, but is closely related to its Nova LLM and Cloud Infrastructure.
These methods are more modular, safer, and are clearly aimed at enterprise use. But they lack the ambition or ambition of autonomy that is shown in Genspark’s presentation.
One reason may be to avoid risks. If a general agent at Google or Microsoft says something weird for the wrong flight or voice call, the reputation cost can be high. These companies are also locked into their own model ecosystem, limiting the flexibility to try to orchestrate multi-models.
In contrast, startups like Genspark can freely mix LLM and move quickly.
Should companies take care of them?
That’s a strategic issue. Most businesses do not need a universal agent to book dinners or make caricatures. However, they may soon need agents that can handle multi-step tasks in specific areas, such as surface and formatting compliance data, planning client onboarding, or producing content across multiple formats.
In this case, Genspark’s work became more relevant. The more seamless and autonomous, the more common proxy they become, the more they integrate voice, memory, and external tools, the more they can start competing with traditional SaaS applications and RPA platforms.
They are using lighter infrastructure to do this. For example, Genspark claims its agents are “super transferable” and are available for marketers, teachers, recruiters, designers and analysts, all with minimal settings.
The age of universal proxy is no longer assumption. It’s here – it’s moving fast.
Watch the video here:
Source link
