Sakana AI Released Sakana Marlin — an Agent for 100-Page Reports with Slides

Tokyo-based AI laboratory Sakana AI on June 15, 2026, presented Sakana Marlin — its first commercial product, a corporate agent that autonomously prepares multi-page research reports with slides.

What Marlin Can Do

The agent works without human participation for up to eight hours on a single task and outputs a ready-made report of up to 100 pages along with presentation slides. For enterprises, this means that part of routine analytical work — data collection, structuring findings, preparing presentations — can be entrusted to a single agent instead of a team of analysts.

Most corporate AI assistants today answer questions in seconds and work within a single dialogue with a human. Marlin is designed differently: the client formulates the task, and then the agent independently works on it for up to eight hours without intermediate prompts, verifying and refining its own conclusions before assembling them into a final document of dozens of pages. This format is closer to the work of an in-house analyst or consulting team than to a dialogue chatbot.

• Launch date — June 15, 2026
• Autonomous operation — up to 8 hours on a single task
• Final report — up to 100 pages with slides
• Based on AB-MCTS technology and the AI Scientist project
• This is the first commercial product in Sakana AI's history

The Technology Behind It

Marlin relies on two internal Sakana AI developments. The first is AB-MCTS (Adaptive Branching Monte Carlo Tree Search), a solution-finding method that the laboratory previously used to help AI models explore and compare multiple alternative lines of reasoning instead of one direct answer. The second is the output from the AI Scientist project, an experimental Sakana AI system that previously demonstrated the ability to independently formulate hypotheses, conduct experiments, and write scientific papers without human researchers.

Marlin essentially transfers this same logic from the world of scientific publications to the world of corporate analytics: the agent explores options, verifies them, and ultimately assembles a coherent document, except instead of a scientific paper, the output is a business report.

Who is Behind Sakana AI

Sakana AI was founded in 2023 in Tokyo by David Ha and Lyon Jones — former researchers at Google Brain and Google DeepMind, respectively. The company's name references the Japanese word for "fish" and reflects the laboratory's approach: instead of one large model, create swarms of small specialized agents that interact with each other like a school of fish. Before launching Marlin, the company was known mainly for research projects like AI Scientist and methods for scaling computations at inference time, such as AB-MCTS, rather than finished commercial products.

Over several years of work, the laboratory has attracted major funding rounds from prominent venture investors and become one of the most prominent AI laboratories in Asia — alongside younger projects from Silicon Valley.

What This Means

The appearance of Marlin demonstrates how research techniques like AB-MCTS quickly transition from laboratory experiments to commercial tools: an agent capable of working autonomously for hours and outputting a document at the level of a consulting report changes expectations for what can be entrusted to AI without constant human oversight. The enterprise-agent sector is growing rapidly: major laboratories are one by one offering tools capable not just of answering questions, but of taking multi-hour tasks through to a finished result. Marlin is Sakana AI's bid for a place in this race, based on its own research methods rather than licensing third-party models.

If the format catches on with corporate clients, the laboratory will have confirmation that its bet on searching through multiple alternative solutions — rather than just increasing model size — works beyond scientific benchmarks.