Join our daily and weekly newsletter for the latest updates and exclusive content on industry-leading AI coverage. learn more
The landscape of a generic AI agent is suddenly much more crowded and ambitious.
This week, a Palo Alto-based startup Genspark I’ve released what it calls Super AgentA fast moving autonomous system designed to handle real-world tasks across a wide range of domains, including several raising eyebrows, such as calling restaurants using realistic composite voices.
The launch adds fuel to what has been formed to become an important new front for AI competitions. Perhaps more urgently, what does that mean for businesses?
The launch of Genspark’s super agent comes just three weeks after another Chinese startup, Manus, we focused on our ability to coordinate tools and data sources to complete asynchronous cloud tasks such as travel booking, resume screening, and inventory analysis.
Genspark now claims to go further. According to co-founder Eric Zinn, the superagent is built on three pillars. Nine different LLM concerts, over 80 tools and over 10 unique datasets all collaborate on a coordinated flow. It handles complex workflows and returns fully executed results, far beyond traditional chatbots.
in demoAgents at Genspark have planned a complete trip to San Diego, calculated walking distances between attractions, mapped public transport mapping options, and used agents to book restaurants with food allergies and seating preferences. Another demo showed that the agent creates a cooking video reel by generating recipe steps, video scenes, and audio overlays. In a third, I riffed on the recent Signalgate political scandal, which involves writing and producing South Park-style animated episodes and sharing war plans with political reporters.
These may sound consumer-oriented, but they show where the technology is heading. This is heading towards multimodal, multi-step task automation that blurs the line between creative generation and execution.
“Resolving these real-world problems is much more difficult than we thought,” Jin says in the video. “But we are excited by the progress we have made.”
One of the persuasive features is that the superagent clearly visualizes its thought process, tracking how it triggers a reason, which tools to invoke, and why. When you see that logic play back in real time, the system no longer feels like a black box, it feels like a co-partner. You can also create similar trackable inference paths for enterprise developers on their own AI systems, making their applications more transparent and reliable.
Super Agent was also impressive to try. The interface started up smoothly in a browser that does not require technical setup. Genspark allows users to start testing without the need for personal credentials. In contrast, Manus still adds friction to the experiment by requiring applicants to join the waitlist and disclose social accounts and other personal information.
We first wrote about Genspark in November when we began financial reporting with Claude. have raised at least $160 million in two roundsand is supported by US and Singapore-based investors.
Please see the latest information Video discussion between AI Agent Developer Sam Witteveen and me To delve deeper into how Genspark’s approach compares to other agent frameworks and why it is important for enterprise AI teams.
How does Genspark bring this out?
Genspark’s approach stands out as it navigates tool orchestration, a long-standing challenge for AI engineering, at scale.
Most current agents break down when juggling just a handful of external APIs or tools. It appears that Genspark’s superagents manage this better by using model routing and search-based selection to dynamically select tools and submodels based on tasks.
This strategy reflects new research into Cotools, a new framework at Soochow University in China that enhances the way LLM uses a wide and evolving tool set. Unlike the old approach, which relies heavily on rapid engineering and rigorous fine-tuning, Kotoll keeps the base model “freeze” and trains smaller components to efficiently judge, get and invoke the tool.
Another enabler is the Model Context Protocol (MCP).which is a lesser known, but increasingly adopted standard, allows agents to carry rich tools and memory contexts throughout the steps. When combined with Genspark’s own dataset, MCP may be one of the reasons their agents appear It is “moblerable” than the alternative.
How does this compare to Manus?
Genspark is not the first startup to promote the general agent. ManusIt was launched last month by China-based company Monica and created waves with a multi-agent system. This completed a multi-step task by autonomously running tools such as a web browser, a code editor, or a spreadsheet engine.
The efficient integration of Manus’ open source components, including web tools and LLMs such as Claude of Manus, was amazing. It outperformed Openai on Gaia Benchmark, despite not building its own model stack. This is a synthetic test designed to evaluate actual task automation by agents.
However, Genspark claims it won 87.8% on Gaia reported by Manus, and 87.8% on Gaia, using an architecture that includes its own components and broader tool coverage.
Big Tech Player: Are you still playing safely?
Meanwhile, the largest AI company in the United States is cautious.
MicrosoftThe main AI agent offering, Copilot Studio focuses on fine-tuned vertical agents that align closely with enterprise apps such as Excel and Outlook. OpenaiThe Agent SDK provides building blocks but stops other than sending its own full features. General purpose agent. AmaZonThe recently announced Nova Act takes a developer-first approach, offering atomic browser-based actions via the SDK, but is closely tied to Nova LLM and Cloud Infrastructure.
These approaches are more modular, safer and clearly targeted towards enterprise use. But they lack shyness in Genspark’s ambitions, or autonomy, at the demonstration.
One reason is risk aversion. If a typical Google or Microsoft agent booking the wrong flight or saying strange things in voice calls, it can lead to higher reputation costs. These companies are also trapped in their own model ecosystems, limiting the flexibility to experiment with multi-model orchestration.
In contrast, startups like Genspark have the freedom to mix and match LLMS and move quickly.
Should businesses care?
That’s the strategic question. Most companies don’t need a general purpose agent to book dinners or produce satirical cartoons. However, they may soon need agents capable of handling domain-specific multi-step tasks, such as surface and formatting compliance data.
In that context, Genspark’s work is more relevant. The more seamless, autonomous, popular agents integrate voice, memory, and external tools, the more they can start to compete with legacy SaaS applications and RPA platforms.
And they do so with lighter infrastructure. Genspark, for example, claims that its agent is “video pilotable” and can be used by marketers, teachers, recruiters, designers and analysts.
A general agent ERA is no longer a hypothesis. It’s here – and it’s moving fast.
Watch the video cast here:
