Genspark's 'Vibe Working' Triples ARR with Rapid AI Product Launches
In an industry often characterized by protracted development cycles and bureaucratic hurdles, AI workspace company Genspark has adopted a radically different approach, which it terms “AI-native working” or “vibe working.” This methodology has allowed the company to accelerate its product releases to an unprecedented pace, launching new features and products almost weekly, and consequently tripling its annual recurring revenue (ARR) growth. Genspark boldly claims this makes it potentially “the fastest-growing startup ever in terms of ARR.”
At the heart of this rapid innovation lies a unique organizational philosophy. Kaihua (Kay) Zhu, co-founder and CTO of Genspark, explains that in an AI-native environment, “basically everybody is the manager.” Individuals are equipped with a team of AI agents, which function as their direct reports, empowering each team member to deliver features end-to-end autonomously. This model contrasts sharply with traditional structures, which Zhu, with over two decades of experience in search at Google and Baidu, believes are prone to friction and inefficiency due to multiple management layers and office politics. Genspark’s lean, 20-person team operates with “less control, more tools,” fostering transparent communication and exceptionally high productivity, where “everybody is working on a product that can ship.”
Launched in June 2024 by MainFunc, Genspark initially focused on AI search, quickly attracting five million users. However, the company soon pivoted to Super Agent, an advanced AI system that dynamically selects the most effective tools and sub-agents for a given task, evaluating results and adapting in real-time. Super Agent, powered by Anthropic’s Claude, debuted on April 2nd and is designed to condense an afternoon of typical office work into mere minutes. Its capabilities span a wide range of functions, from making calls and performing deep research to fact-checking, drafting documents, producing podcasts, and generating spreadsheets and presentations.
The impact of this “gen speed” development model is clearly visible in Genspark’s aggressive rollout schedule and financial milestones. Just nine days after Super Agent’s launch, on April 11th, the company reached $10 million in ARR. This momentum continued, with new features like AI Slides (April 22nd) and personalized Super Agents (April 28th) quickly following. By May 2nd, exactly one month post-launch, Genspark’s ARR had surged to $22 million, further climbing to $36 million by May 19th. The subsequent months saw a continuous stream of innovations, including AI Sheets (May 8th), an agentic download system and AI drive (May 15th), AI-powered phone calls (May 22nd), an AI Secretary for managing communications and calendars (June 4th), and an AI Browser with an extended tool marketplace (June 10th). The relentless pace continued through July and into August, with the introduction of AI Docs, Design Studio, AI Pods for podcast creation, and finally, multi-agent orchestration, enabling up to ten AI agents to operate simultaneously.
Genspark’s rapid ascent has also fueled a competitive spirit within the burgeoning AI agent space. Following OpenAI’s mid-July announcement of its ChatGPT agent, Genspark initiated a “1 Million Dollar Side-by-side AI Showdown,” challenging users to identify instances where other platforms outperform Super Agent. In the first round, users were tasked with creating a 12-page financial slide using both Genspark and ChatGPT Agent, with 429 instances found where the latter edged out, earning participants $100 for each. The second round, which concluded in early August, upped the stakes to $200 per win and broadened the competition to any AI tool, with results evaluated by Google Gemini. Genspark framed this contest not as a rivalry but as a collective effort to push the boundaries of the AI agent ecosystem.
The technical underpinnings of Super Agent are as sophisticated as its development process. Unlike older search paradigms that relied on rigid, fixed workflows, Super Agent employs a mixture-of-agents (MoE) system, integrating nine distinct large language models (LLMs) of varying sizes and specializations. These models collaboratively break down tasks, delegate responsibilities based on individual strengths, and cross-verify one another’s outputs. Super Agent is also equipped with over 80 tools, ranging from sub-agents capable of generating Python code to those that can autonomously make phone calls, and draws upon more than 10 curated datasets. Genspark leverages a diverse array of foundational models, including those from Anthropic, OpenAI, Google Gemini, DeepSeek, and xAI’s Grok 4, using an aggregator model to analyze their outputs for optimal cost-effectiveness, accuracy, and reduction of “hallucinations.” While Genspark also fine-tunes its own frontier model, Zhu emphasizes that the goal is not to chase state-of-the-art breakthroughs for their own sake, but rather to reduce costs and latency for high-volume, lower-level tasks, as many proprietary models are “too big, too slow and too expensive.”
Ultimately, Genspark’s “vibe working” ethos extends beyond just engineering, aiming to democratize AI development. The company believes that by making its tools intuitive and powerful, it can enable even non-programmers to “vibe”—experiment and create with AI—without needing familiarity with complex integrated development environments or coding languages.
Genspark’s AI-native approach isn’t just accelerating product cycles; it’s redefining how software is built and scaled.