Recently, the startup MainFunc, headquartered in Palo Alto, California, launched the so-called Genspark Super Agent, an autonomous system capable of rapid response, independent thinking, planning, acting, and using tools to assist users with all daily tasks.So, what is this AI Agent, and what can it specifically do?The company behind Genspark Super Agent, MainFunc, was founded at the end of 2023 by Eric Jing and Kay Zhu. Eric Jing previously served as CEO of Baidu’s Xiaodu Technology, Vice President of Baidu Group, General Manager of Baidu’s Smart Life Group (SLG), and Chairman of Baidu’s AI Product Committee. Before joining Baidu, he was the Chief R&D Director at Microsoft, responsible for R&D in the Asian market for Bing, and was also the creator of Microsoft’s Xiaoice.Kay Zhu served as CTO of Xiaodu Technology and was previously a Principal Architect at Google, where he was the creator of core algorithms such as Google Panda.The team consists of about 20 senior members from Microsoft, Google, and Baidu, with headquarters in Palo Alto, California, and Singapore, and plans to open a new office in Seattle.The Genspark Super Agent they released is a new type of general AI Agent designed to autonomously handle complex tasks across various domains. Unlike simple chatbots or scripts, the Genspark Super Agent can “think, plan, act, and use tools” like a human assistant. It can not only generate text but also take actions on behalf of the user. When users provide high-level instructions (similar to project briefs or SOPs), it will internally break down the problem, formulate a plan, and execute it step by step with minimal supervision. This means it can perform multi-step tasks such as researching information, converting data, and even executing real-world actions (like making phone calls) without continuous guidance.Architecturally, the Super Agent features a unique design that employs a “hybrid agent” approach, integrating eight different large language models (LLMs), over 80 internal tools, and dozens of curated datasets. Depending on the complexity, speed, and accuracy requirements, each task is intelligently routed to the best model, ensuring efficient and precise execution.The Story Behind GensparkIn 2024, MainFunc launched its first product, Genspark, the “AI Agentic Engine,” positioned as an AI Agent engine focused on providing search services. With its innovative concept, Genspark quickly rose to the fourth position on Product Hunt’s daily new products list.In June 2024, MainFunc completed a $60 million seed round of financing led by Singapore’s BlueRun Ventures, bringing the company’s valuation to $260 million. This funding will accelerate the development and market expansion of Genspark.In February of this year, MainFunc was reported to have raised $100 million in Series A funding, with a valuation of $530 million.The attention Genspark has received is due to its nature as not just a search engine but an AI agent engine aimed at leveraging AI to provide a high-quality search experience. It addresses many shortcomings of traditional search engines, such as endless ads, biased information, and a plethora of irrelevant results.The innovation behind Genspark lies in its unique “Sparkpages” concept. Sparkpages are custom pages generated by professional AI agents that provide fair and reliable information, free from SEO-driven content interference. These pages save users valuable time by delivering more meaningful and valuable search results.After accumulating 5 million users, the company pivoted to Super AgentCurrently, Genspark primarily targets the U.S. market, offering its services for free, with plans to introduce a paid subscription model in the future.Recently, MainFunc’s CTO Kay Zhu stated in a letter released on Product Hunt: “We made an intuitive decision—despite our AI search product covering over 5 million users, we have decided to abandon it. The reason is that we believe traditional AI search is outdated.”Kay Zhu shared the company’s experience over the past nine months in building this Super Agent.
In a demonstration by Eric Jing, the Genspark Super Agent planned a five-day trip to San Diego, calculated walking distances between attractions, mapped public transport options, and then used a voice call agent to book restaurants, including handling food allergies and seating preferences. Another demonstration showcased the agent creating a cooking video reel by generating recipe steps, video scenes, and audio overlays.
These may sound consumer-focused, but they showcase the direction of technological development—multi-modal, multi-step task automation that blurs the lines between creative generation and execution.“Solving these real-world problems is much more challenging than we imagined,” Eric Jing said in the video, “but we are very excited about the progress we have made.”A notable feature: the Super Agent clearly visualizes its thought process, tracking how it reasons through each step, which tools it calls upon, and why. Observing the execution of logic in real-time makes the system feel less like a black box and more like a partner. It can also inspire enterprise developers to build similar traceable reasoning paths in their own AI systems, making applications more transparent and trustworthy.Predefined workflows cannot scaleWhen Perplexity debuted at the end of 2022, it sparked interest and expectations regarding AI’s potential to disrupt search. This was because previous search queries followed a strict workflow: analyzing queries and expanding keywords; retrieving top-ranked web results; reordering/summarizing them into final responses.While this framework is sufficient for basic issues, it clearly falls short for more complex needs, such as comparing technical solutions, planning multi-factor procurement decisions, or handling in-depth research.At Genspark, the team initially built their AI search engine using the same process, then made gradual improvements:Utilizing specialized data sources (academic, financial, travel, product, etc.) to enrich information content. Automatically triggering in-depth investigations for complex queries through parallel searches. Cross-checking asynchronous agents to validate statements that are too complex to process quickly, then further expanding into data search and deep research agents. Adopting a hybrid agent approach to combat hallucinations, with multiple agents verifying each other’s outputs.Kay Zhu stated that while these innovative methods significantly improved search quality and expanded the user base, they realized that the entire design was still constrained by traditional thinking, adhering to fixed and predefined workflows. To truly establish a solution with strong adaptability and rich context, the team had to break free from these constraints. This realization led them to create the Super Agent—a new paradigm that integrates all past improvements without being bound by static step sequences.Engineering adaptability design in Super AgentThe Genspark Super Agent no longer forces every query through a fixed workflow but flexibly adjusts based on the problem at hand. It plans each step, uses the best tools or sub-agents, observes results, and adjusts strategies in real-time, often displaying surprising creativity. If one approach fails, the Super Agent smoothly transitions to another method, scaling to simulate human problem-solving methods.This flexibility is evident both in breadth (i.e., from which data sources or APIs to extract data) and in depth (i.e., how many rounds of iterative improvements are made). For simpler tasks, it does not waste time on unnecessary steps; for more complex tasks, it can continue digging until a satisfactory answer is found. It can also customize its output results based on each user’s needs—including direct answers, Sparkpages (articles), presentations, generated images, interactive HTML pages, and even voice calls.Large models, tools, and curated data in a trinity
Dynamic orchestration and model guidanceThe Genspark Super Agent coordinates eight specialized large models through a hybrid agent framework. Each model is pre-guided to ensure excellent performance in its assigned role, covering areas such as rapid response and in-depth analysis. This hierarchy ensures stability while maximizing the unique advantages of each model.Dedicated tools and sub-agentsThe technical team has pre-designed a library of sub-agents covering various components, from presentation generators and Python code executors to calling modules. Each sub-agent is optimized for reliability and efficiency, ensuring the Super Agent can handle tasks like chart creation and interactive page development without users being constrained by rigid workflows.Reliable curated dataThe Super Agent accesses carefully validated datasets sourced from high-quality web sources, trusted partners, and expert-reviewed repositories. Offline validation agents continuously audit and perfect this data. By prioritizing accuracy over quantity, misinformation is minimized, ensuring the reliability of outputs.Lessons learned: Reduce control, increase toolsAs the transition from rigid AI search engines to a more fluid and adaptive Super Agent occurred, the Genspark team discovered a fundamental principle: reduce control, increase tools. Overly structured workflows can stifle creativity and functional depth, while introducing multiple specialized agents to address different aspects of a problem and granting them the freedom to choose and switch between various tools can unlock greater capabilities.Reduce control:Prescriptive processes often limit the scope of exploration, making systems harder to adapt to new challenges. Adopting a more open strategy helps foster innovation and resilience.Increase tools:Equipping agents with specialized retrieval, analysis, presentation, and communication modules enables them to devise end-to-end solutions in real-time. This not only supports advanced use cases but also ensures task lightweight and execution efficiency.This exploration, which integrates adaptive, diverse tool support with vetted data, makes the Genspark Super Agent more flexible and reliable than any traditional AI search product. Faced with such significant advantages, the Genspark team resolutely decided to abandon their already thriving AI search solution to focus on the new direction represented by the Super Agent.
Benchmarking Against ManusSo, how does Genspark compare to other AI Agents?By extensively integrating models and directly using APIs, Genspark distinguishes itself from competitors like Manus AI and OpenAI Operator. While Manus AI primarily relies on fewer models and offers fewer integrated tools, Genspark’s broader toolset and real-world interaction capabilities (including voice calls) provide additional practical advantages. Compared to the OpenAI Operator, which is limited to browser-based operations, Genspark’s API-centric approach allows for faster, more structured data retrieval and richer task execution capabilities.
Industry experts point out that the multi-agent design of the Super Agent represents an important trend in AI Agent technology. Compared to traditional single-model systems, this hybrid architecture not only enhances task processing flexibility but also significantly reduces error rates and the probability of “hallucinations” (producing inaccurate content). For instance, in the GAIA benchmark tests, the Super Agent outperformed some OpenAI models and Manus, demonstrating its high reliability in automating task execution. This advantage is particularly pronounced in scenarios requiring high accuracy and controllability.However, the launch of the Super Agent also raises some unresolved questions. Some commentators have pointed out that the specific implementation details of its multi-agent system have not been fully disclosed, and its performance in extremely complex tasks or resource-constrained environments still needs further validation. Additionally, as AI agent technology becomes more widespread, ensuring data privacy and system transparency will be challenges that Genspark needs to address in the future.The Genspark Super Agent allows users to start testing without personal credentials.Reference Links:https://www.marktechpost.com/2025/04/05/meet-genspark-super-agent-the-all-in-one-ai-agent-that-autonomously-think-plan-act-and-use-tools-to-handle-all-your-everyday-tasks/https://venturebeat.com/ai/gensparks-super-agent-ups-the-ante-in-the-general-ai-agent-race/https://mainfunc.ai/blog/genspark_kill_aisearchhttps://mainfunc.ai/blog/genspark_moa_powered_searchhttps://www.producthunt.com/posts/genspark-super-agent
Disclaimer: The purpose of this public account (ID: cnzifi) in publishing content and images is to disseminate more information, and the copyright belongs to the original author. It is not for commercial use. If there is any infringement, please contact us.
