Sakana AI launches Sakana Marlin, an agent that produces 100-page reports with slides

Tokyo-based AI lab Sakana AI unveiled Sakana Marlin on June 15, 2026. It is the company's first commercial product: an enterprise agent that autonomously produces multi-page research reports with slides.

What Marlin can do

The agent works without human involvement for up to eight hours on a single task and delivers a finished report of up to 100 pages, complete with presentation slides. For enterprises, this means a single agent can take on part of the routine analytical work a team of analysts would otherwise handle: gathering data, structuring findings, and building the presentation.

Most enterprise AI assistants today answer questions in seconds and operate within a single conversation with a human. Marlin works differently. The customer defines the task, and the agent then works on it independently for up to eight hours without intermediate prompts, checking and refining its own conclusions before compiling them into a final document dozens of pages long. The format looks more like the work of an in-house analyst or a consulting team than a conversational chatbot.

• Launch date: June 15, 2026
• Autonomous operation: up to 8 hours per task
• Final report: up to 100 pages, including slides
• Built on: AB-MCTS and work from the AI Scientist project
• This is Sakana AI's first commercial product in the company's history

The technology behind it

Marlin draws on two of Sakana AI's in-house developments. The first is AB-MCTS (Adaptive Branching Monte Carlo Tree Search), a solution-search method the lab previously used to have AI models explore and compare several alternative reasoning paths instead of producing a single direct answer. The second is work from the AI Scientist project, Sakana AI's experimental system that previously demonstrated it could formulate hypotheses, run experiments, and write scientific papers on its own, without human researchers.

In essence, Marlin carries the same logic from scientific publishing over to corporate analytics. The agent likewise explores options, checks them, and ultimately assembles a coherent document, except the output is a business report rather than a scientific paper.

Who is behind Sakana AI

Sakana AI was founded in Tokyo in 2023 by David Ha and Llion Jones, formerly researchers at Google Brain and Google DeepMind, respectively. The company's name comes from the Japanese word for "fish" and reflects the lab's approach: rather than building one large model, it creates swarms of small, specialized agents that interact with each other, much like a school of fish. Before Marlin, the company was known mainly for research projects such as AI Scientist and for inference-time compute scaling methods such as AB-MCTS, not for finished commercial products.

In a few years of operation, the lab has raised large funding rounds from well-known venture investors and become one of Asia's most prominent AI labs, alongside younger projects from Silicon Valley.

What it means

Marlin shows how quickly research techniques like AB-MCTS are moving from lab experiments to commercial tools. An agent that can work autonomously for hours and deliver a document on par with a consulting report changes expectations about what can be handed off to AI without constant human oversight. The enterprise agent sector is growing fast, with major labs releasing one tool after another that can do more than answer questions: they can carry multi-hour tasks through to a finished result.

Marlin is Sakana AI's bid for a place in this race, built on its own research methods rather than on licensing other companies' models. If the format catches on with corporate customers, the lab will have proof that its bet on searching across many alternative solutions, rather than simply scaling up model size, works beyond scientific benchmarks.