Sakana AI launches Marlin to automate in-depth strategy research

Most AI tools try to give you an answer in five seconds flat, usually resulting in surface-level corporate fluff that leaves you doing the real research yourself. Sakana AI is taking the exact opposite approach: letting the model think for hours.

Tokyo-based Sakana AI launched Sakana Marlin, its first commercial product. Pitching it as a virtual Chief Strategy Officer, the company built Marlin to tackle the heavy strategic research that usually takes a CSO and a small team weeks to pull off.

Eight hours of solo deep-dive research

The workflow starts with a brief exchange. You give Marlin a research topic, and it asks a few questions to sharpen the scope and align on direction. Once you give the green light, humans step aside.

From there, Marlin works autonomously for up to roughly eight hours. Instead of running a single web search, it continuously generates hypotheses, tracks down data, verifies findings, and maps out causal relationships in complex business environments.

When it finishes, you get a full strategic report up to a hundred pages long, paired with executive-ready summary slides. The goal is to skip basic summarization and hand management structured strategic options they can actually decide on.

To get this enterprise-ready, Sakana ran a closed beta starting in April 2026 with around 300 professionals from investment firms, corporate strategy divisions, consulting outfits, and think tanks. Under the hood, Marlin combines Sakana’s long-horizon reasoning research, including techniques from its Nature-published The AI Scientist and its NeurIPS Spotlight multi-model coordination framework AB-MCTS. The tool is available now via a free pay-per-use tier alongside Pro, Team, and Enterprise plans, though regional availability restrictions apply.

Why it matters

We think Sakana made a very clever bet here: competitive analysis and strategy decks are massive profit centers for consulting firms, yet most of the hours go into grunt-work desk research. For strategists, consultants, and founders, Marlin offers a digital junior team that will grind through hundreds of sources while you sleep. If your job involves writing comprehensive industry teardowns or market entry strategies, taking the pay-per-use tier for a spin is a no-brainer.

For the AI landscape, Marlin signals an important shift from instant chat answers to long-horizon, autonomous execution. Most business problems cannot be solved in a 30-second context window. By letting reasoning run for up to eight hours across coordinated models, Sakana is leaning into the idea that real intelligence takes deliberate time and multi-step verification.

The clear risk, however, is error compounding. When an agent runs unsupervised for eight hours, an early factual slip or misframed assumption can cascade across a hundred-page document. Executive teams cannot afford polished hallucinations. While generating massive slide decks autonomously is an engineering feat, the real test will be whether executives find genuinely contrarian insights inside those hundred pages, or just very structured, beautifully formatted noise.