Fireworks AI’s Lin Qiao wants to make AI agents cheaper to run

AI-powered apps can’t always afford to serve their customers. “Product market fit does not equal a viable business,” says Lin Qiao, CEO of Fireworks AI. “If startups aren’t careful, they can scale into bankruptcy.” Her company gives developers access to high-performance chips and the infrastructure to run artificial intelligence models in production, helping more than 10,000 companies, including Cursor, Uber, and DoorDash, build agents and workflows. Lin sees 2026 as the year AI agents move from splashy demos to being used all over the place. She is already seeing it happen inside her own company: The Fireworks finance team automated its reporting and planning, the legal team built contract review tools, and the recruiting team uses AI agents for candidate sourcing. (Even an Uber driver Lin met moonlights as a music producer by using AI to write lyrics and generate tracks.) On Fireworks AI’s inference platform alone, more than 15 trillion tokens are processed each day, making agents enormously expensive to operate. Lin is pioneering a different approach: smaller, customized models trained on each application, which can boost efficiency 5 to 10 times. “Every single use case should have its own continuously customized model,” she says. Fireworks has scored a $4 billion valuation. “This isn’t just about companies building AI,” Lin says. “It is about individuals automating their work.” Before founding Fireworks, Lin cocreated PyTorch, the Meta open-source framework underpinning many AI models today. That helped her come to the conclusion that the future belongs not to one dominant model, but to millions of specialized ones. She’s confident that’s viable because open and closed models are quickly converging. “Every two weeks, there’s a new model launch. Closed, open, doesn’t matter. And they compare with each other, right?” Lin says. “Each model launch, they have to top some kind of leaderboard. Otherwise, they feel it’s not even worth launching. Whenever you see oscillation, that just means it’s converging. And we start to see a lot more people open to using open models, because the quality is getting much better.” This profile is part of Fast Company’s AI 20 for 2026, our roundup spotlighting 20 of AI’s most influential technologists, entrepreneurs, corporate leaders, and creative thinkers. The extended deadline for Fast Company's Next Big Things in Tech Awards is Thursday, June 18, at 11:59 p.m. PT. Apply today.