Data Infrastructure, Not AI Models, Limits Agent Success

Meta is deploying its in-house coding agent MetaCode to thousands of engineers to improve the coding capabilities of its AI models and close the gap with Anthropic and OpenAI. VP Maher Saba has asked engineers to submit at least one code change per week for review and integration. The feedback loop has already improved Meta's latest model, Muse Spark 1.1, and will be used to train an upcoming model called Watermelon.

AI is accelerating drug discovery by enabling predictive design of candidates and hit identification at scale, but the technology is exposing critical gaps in data quality and lab infrastructure. Drug companies are hitting a 'data wall' where publicly available datasets lack the structure and diversity needed to train accurate models, while lab teams struggle to validate the growing volume of AI-generated compounds. Success depends on closing the loop between computational prediction and experimental validation through better data collection and integration.

Frontier physical AI models are moving beyond video training data to incorporate multiple camera angles, dense annotation, and brain wave readings as training inputs. The shift reflects growing recognition that traditional video datasets alone are insufficient for training AI systems that interact with the physical world. Brain wave data represents an emerging frontier in multimodal training approaches for robotics and embodied AI.

Mercor, a three-year-old data startup that trains AI models through contractor networks, generated $614 million in gross revenue in the first half of 2026, up 70% from all of 2025. The company's growth is heavily concentrated among AI foundation model makers, with 91% of first-half revenue coming from customers like OpenAI, Anthropic, and Google DeepMind. This revenue concentration reveals both the startup's market traction and its dependency on a narrow customer base.