Sakana AI Launches 'Marlin,' an Autonomous Research Agent That Thinks for Hours Instead of Seconds

Tokyo-based artificial intelligence startup Sakana AI has officially launched its first commercial product, an autonomous business-to-business research agent named Sakana Marlin. Billed by the company as a "Virtual Chief Strategy Officer," the platform is designed to take on the heavy lifting of comprehensive corporate research that would typically require weeks of human effort. The launch marks a significant milestone for the company, which was founded in 2023 by former Google researchers and has quickly gained a reputation for its unorthodox approaches to AI architecture. Available immediately for enterprise customers, Marlin targets corporations, financial institutions, and think tanks that require deep, rigorous analysis rather than quick conversational answers.[1][5][7]

Marlin represents a deliberate and striking departure from the instantaneous text generation that has defined the generative AI boom over the past two years. The industry standard has largely been defined by speed—the ability to generate a poem, a line of code, or a surface-level summary in mere milliseconds. Instead of returning a rapid-fire reply, Marlin is built from the ground up for deep, long-horizon reasoning. With this launch, the enterprise frontier is rapidly shifting toward methodical, extended thought processes where accuracy and depth take precedence over immediate gratification. Major businesses are no longer asking how fast an AI can answer a query, but rather how deeply it can think about a complex strategic problem.[1][2]

The workflow for Sakana Marlin fundamentally alters how professionals interact with large language models, requiring minimal human intervention once a task is initiated. Rather than engaging in a tedious, back-and-forth prompt engineering session, a user simply provides a core research topic. Following a brief initial exchange to sharpen the scope and direction of the investigation, the human steps away entirely. For up to eight hours, Marlin operates as a self-contained digital strategy team. It does not require a human to guide it through intermediate steps, ask follow-up questions, or correct its course, making it a truly autonomous agent rather than a simple digital assistant.[1][3][7]

During this extended eight-hour runtime, the agent autonomously cycles through a rigorous scientific process. It formulates its own initial hypotheses, navigates the live web to gather raw data, cross-references multiple sources, and actively works to resolve internal contradictions. If the AI encounters a dead end or conflicting information, it is designed to recognize the discrepancy and adjust its search parameters accordingly. This continuous, self-governing reasoning loop allows Marlin to map the causal relationships at work in complex business environments and organize them into structured strategic options, mimicking the iterative research process of a human analyst.[1][7]

The final output produced by Marlin is far more substantial than a standard chatbot response, reflecting the hours of computational effort invested in the query. The system delivers structured executive presentation slides alongside a deeply researched, fully cited strategy report that can span up to 100 pages. At a recent hands-on demonstration for the media in Tokyo, Sakana AI showcased reports generated by Marlin on complex, real-world topics such as the geopolitical risks of a new US administration and the impact of supply chain disruptions. Reporters reviewing the outputs noted the sheer volume and depth of the generated documents, with individual reports containing between 60 and 80 distinct source citations.[4][5]

The underlying technology driving Marlin relies on a concept known as "inference scaling." Rather than relying solely on a massive, pre-trained model to guess the next word based on historical data, Marlin spends significantly more computational power during the actual reasoning phase. This is powered by an algorithm Sakana AI calls Adaptive Branching Monte Carlo Tree Search (AB-MCTS). The system treats the complex research process like navigating a branching tree of possibilities. It continuously evaluates the validity of each hypothesis, concentrating its computational resources on the most promising analytical routes while discarding dead ends, proving that longer "thinking" time directly yields higher-quality insights.[2][5][7]

Furthermore, Marlin integrates crucial findings from the company's "AI Scientist" project—a framework originally designed to automate the entire scientific discovery process from idea generation to peer review. By combining these advanced technologies, Sakana AI has created a system where multiple specialized AI models work cooperatively to complete the workflow. One model might focus on data retrieval, while another handles logical structuring, and a third verifies the citations. This multi-agent approach allows Marlin to tackle multifaceted strategic problems that would overwhelm a single, monolithic language model, bringing a level of academic rigor to corporate market research.[2][6][7]

Prior to its official commercial launch on June 15, 2026, Sakana Marlin underwent a rigorous closed beta testing phase that began in April. Approximately 300 professionals from highly regulated and demanding sectors utilized the tool for market research, risk analysis, and competitive intelligence. Testers from major financial institutions, consulting firms, and think tanks provided feedback that helped refine the agent's output formatting and long-term stability. The successful deployment in these strict corporate environments—including partnerships with institutions like Sumitomo Mitsui Financial Group—serves as a strong proof of concept for the reliability of Sakana AI's autonomous architecture.[5][6]

The promise of compressing weeks of strategic planning into a single workday is highly attractive to enterprises looking to optimize their operations. Proponents argue that by outsourcing the grueling, time-intensive data-gathering phase to an autonomous agent, human executives can focus their energy entirely on high-level decision-making and nuanced judgment calls. Instead of spending weeks compiling data on a new market entry, a Chief Strategy Officer can arrive at work, review a comprehensive 100-page brief generated overnight, and immediately begin debating the strategic merits of the proposed options with their board of directors.[3][7]

However, the shift toward long-running autonomous agents also introduces new operational challenges and requires a high degree of trust. Industry observers and risk analysts point out that allowing an AI to run hundreds or thousands of self-directed loops over an eight-hour period makes it difficult to spot subtle errors or factual hallucinations that might be buried deep within a massive report. To mitigate these concerns, Sakana AI has emphasized transparency, ensuring that Marlin's outputs are meticulously cited so users can trace the agent's logic back to the original sources. Ultimately, the launch underscores a broader pivot in the tech sector: the next major frontier lies in autonomous agents capable of executing complex, multi-step enterprise workflows.[1][2][6]