For most of the last decade, progress in AI has followed a simple rule: scale. Bigger models, more data, more compute. When something didn’t work, the answer was usually to make it larger.
That approach delivered remarkable breakthroughs, but it also created systems that are rigid, costly, and difficult to use in the real world. Training frontier models has become increasingly expensive, roughly 2.4x per year since 2016, with future runs projected to exceed $1 billion by 2027.
Most AI founders we've met in the last year have pitched new labs that followed the same scale-at-all-costs approaches as their predecessors. When we met Sara Hooker and Sudip Roy, we quickly realized that they build to a different tune. Adaption Labs was founded around a firm conviction: scale alone is no longer enough.
Sara and Sudip reached this insight after years at the front lines of production at Google and Cohere. Sara led cutting-edge research while building systems that actually run in the real world. Sudip engineered infrastructure that delivers reliable, scalable performance. In that work, they saw firsthand where models strain under real workloads. That experience revealed a simple truth: AI that works in practice can’t stay frozen. It must adapt and continuously learn.
Builders Who Saw the Limits Firsthand
Sara is a global leader in AI whose work spans the full stack, from pretraining to production, with a proven track record of building systems for diverse environments. She has led research at Google Brain and Cohere, advancing AI across hundreds of languages and building systems that perform reliably under diverse, real-world conditions. Recognized as one of the most influential voices in the field, Sara combines technical mastery with deep commitment to creating AI that is responsible and inclusive, serving on advisory councils shaping the future of the field.
Sudip brings a rare systems-first foundation to AI. He builds intelligence that is not only powerful but also fast, reliable, and capable of thriving in the real world. At Google, he worked on core infrastructure like Pathways, and on systems that serve models such as Gemini in production. At Cohere, he led inference and platform efforts, working closely with hyperscalers and enterprise customers. He knows that intelligence that falters under pressure can’t form the foundation of a business.
That understanding matters more now than ever. Enterprises are no longer experimenting at the margins. Across more than 10,000 organizations, enterprises are putting roughly 11x more models into production year-over-year, signaling a clear shift from experimentation to operational systems. As AI moves deeper into real workflows, reliability and cost control stop being nice-to-haves.
What makes Sara and Sudip’s partnership special is not just complementary backgrounds. It is how they reimagine building intelligence together. Research, systems, and interface design aren’t sequential steps or afterthoughts. They’re co-designed, in lockstep, from the ground up.
Most AI systems today are fixed at deployment. A model is trained, released, and expected to handle every request the same way. When it fails, users adapt instead. Prompts grow longer. Workflows become fragile. Costs rise as systems are forced to reason deeply even when the task is simple.
Sara and Sudip invert that dynamic.
A Different Way to Build Intelligence
Adaption is building a new kind of intelligence that responds to user intent and adapts to the context of each task. Rather than forcing every problem through a single general-purpose model, their systems adjust dynamically, evolving over time.
In practice, that means not every request is treated equally. Simple tasks remain lightweight. More complex work draws on deeper reasoning only when necessary. The system adapts across data, models, compute, and interface to match each workload. Intelligence becomes faster, more reliable, and significantly more efficient because it is not doing unnecessary work.
This design choice transforms the economics of AI. Cost and latency scale with task complexity, not model size. For developers and enterprises running repeatable workloads, that difference determines whether AI stays experimental or is pushed to real-world production use cases.
In an industry still obsessed with brute-force scaling, Adaption takes a different path: prioritizing efficiency and adaptability over size alone. That belief is rooted everywhere, from the architecture of the platform to the team the founders are assembling.
Changing the Way the World Works
At Emergence, we aspire to work with founders who build enduring businesses with lasting impact at a global scale. What stands out is rarely a single insight. It is a pattern of judgment that shows up early in what founders choose to prioritize and the scale of the impact they aim to contribute.
It was clear from early conversations that Sara and Sudip’s frontier backgrounds and breakthrough research contributions have given them a unique lens into how AI systems should work in the future. But, even more importantly, it was clear that their ambitions were limitless.
That is why we are thrilled to be leading Adaption’s $50 million seed round.
This investment represents our conviction in the next phase of global AI. A shift away from static intelligence toward systems that learn from use and progress measured by reliability and real-world impact.
Adaption is introducing itself in stages, starting with a clear vision for adaptive AI systems. The goal is to put adaptable systems into the hands of developers and businesses, letting real usage shape what comes next.
Every intelligent system we know adapts. It would be strange if AI did not.
That is the shift Adaption is setting in motion.
.jpg)
