BLOG
We launched with a tiny Japanese hotel for a simple reason. Access. Intelligence should not be a luxury. It should help the smallest operator show up with the same clarity as the largest brand.
Google's Gemini 3.0 Flash delivers measurably better accuracy than 2.5 Flash, and that improvement compounds at scale. We tested it on complex multi-step navigation—the results were immediate: faster execution, more precise decisions, cleaner output. For production workflows, a 5% accuracy gain translates to hundreds fewer failures per month. If consumer browser agents felt too inconsistent for production, this changes the equation.
The internet gave humans access to knowledge. Now AI agents need access to action. The web wasn't built for machines. It was built for humans with browsers and clicks. TinyFish is building the infrastructure that makes the web operable for agents: stable contracts instead of brittle DOMs, outcomes instead of search results, reliable execution at scale. Google organized the world's information. We're making it executable.
A procurement agent selects a faulty supplier. A fleet robot chooses contaminated fuel. Current liability frameworks assume human judgment in the loop. Courts will soon discover that assumption no longer holds.
Urgency timers, hidden fees, confirmation tricks, decades of conversion optimization exploited human cognitive bias. Machines parse terms programmatically. The manipulation toolkit becomes suddenly obsolete.
Automation is not only imposed; it is often welcomed. The exhausted consumer delegates decisions willingly. Why we surrender choice may matter more than the technology that enables surrender.
Checkout aisles, upsells, and last-minute additions depend on human irrationality. When agents and robots execute against constraints, entire margin structures collapse.
Every day, Mino navigates thousands of websites while passing tests designed to prove something impossible: that this enterprise web agent is human.
A conversation between TinyFish co-founder Shuhao Zhang, CTO Mark Peng, and OpenAI's Dmitry Pimenov focusing on the evolution of AI agents technology.
Search engines index less than 5% of online content, with the remaining 90%+ existing behind authentication walls, interactive workflows, and dynamic interfaces.
Enterprises require systems functioning reliably under production conditions, not merely functional demonstrations. TinyFish proposes "codified learning" as a solution.
The web has fundamentally transformed from a simple searchable index into a complex system where information is concealed behind authentication, dynamic content, and personalized experiences.
Live demo of TinyFish agents navigating real websites, handling edge cases, and completing complex multi-step workflows.
How we're partnering with OpenAI to enable GPT models to take action across the entire web.
The hidden web that APIs can't reach—and how TinyFish unlocks it for AI agents.
How TinyFish agents learn from every interaction to become more reliable, faster, and smarter over time.
Series A funding led by a16z to accelerate the agentic internet economy.
