The age of building better models is over. Now, the smartest minds in AI are asking a different question: What if the next breakthrough isn’t in the algorithm, but in the data itself?
At the AI Agent Conference in New York City, industry leaders and AI practitioners like Robert Nishihara, co-founder of Anyscale, gathered to explore how artificial intelligence is evolving beyond models and moving into a new era driven by data innovation.
The conversation centered around three transformative paradigm shifts in AI development that redefine how we think about performance, training, and intelligence.
1. AI Is Now Data-Centric
One of the summit’s core takeaways was a clear pivot from model-centric to data-centric innovation. As learning algorithms become increasingly standardized and pre-training gains level out, the competitive edge is now found in data quality itself.
Experts discussed how the frontier of AI is shifting toward compute-intensive processes focused on generating higher-quality, optimized datasets. This dynamic approach to data generation will filter out low-quality inputs and leverage AI to improve and refine the data lifecycle—moving from a static to a dynamic data model.
2. The Rise of Unsupervised Learning
A decade ago, the idea of teaching machines with unlabeled data felt out of reach. Today, unsupervised learning has matured from a theoretical concept to a practical tool. Summit speakers explained how AI now leverages unlabeled inputs to predict outcomes—like using the first ten words in a sentence to predict the eleventh, a mechanism familiar to anyone who uses email auto-complete. This evolution enables AI systems to learn patterns and structures without the need for costly and time-intensive human labeling, marking a major leap in scalability and autonomy.
3. Compute as the Engine of Smarter AI
The path to smarter models is no longer just about fine-tuning neural networks—it’s about where and how we apply compute power. By investing in computing to create better training data, AI systems can self-generate, evaluate, and refine data sets at scale. The result? An upward spiral of quality enhances the accuracy of models and their ability to adapt in real time.
For Amy Cook, Fullcast co-founder and CMO, who emceed the event, the Agentic AI Summit spotlights a major inflection point in the AI landscape. As models stabilize and mature, it’s the data—how it’s generated, curated, and dynamically improved—that’s becoming the real frontier. Unsupervised learning and compute-intensive data generation are no longer experimental—they’re the foundation of AI’s next chapter.
