The Ukrainian conflict demonstrates the power of a fast, iterative cycle: deploy technology, see if it works, and adapt quickly. This agile approach, common in startups but alien to traditional defense, is essential for the U.S. to maintain its technological edge and avoid being outpaced.

The strategy's focus on AI simulation acknowledges a key risk: AI systems can develop winning tactics by exploiting unrealistic aspects of a simulation. If simulation physics or capabilities don't perfectly match reality, these AI-derived strategies could fail catastrophically when deployed.

Overly prolonged lab test and evaluation cycles prevent AI capabilities from ever experiencing real-world field conditions. Because enemies, environments, and warfighters dynamically alter operational realities, pre-deployment models cannot be tuned to absolute perfection in advance. Teams must bound evaluation to ensuring basic safety and understanding risks, deploying models into the field quickly so real exposure drives adaptation.

Fielded autonomy in modern warfare, as observed in Ukraine, did not begin with autonomous vehicles; it emerged because technicians spent years actively collecting, transmitting, and annotating video and sensor streams from human operators. In contrast, the US military lacks continuous data collection pipelines for machine learning, frequently discarding or siloing exercise data due to storage costs and institutional neglect of boring data infrastructure.

Instead of perfecting AI in a lab, Project Maven deliberately deployed flawed, early-stage systems to frontline operators. They accepted initial user frustration and system failures as a necessary cost to gather real-world feedback and rapidly iterate, a stark contrast to traditional, slow-moving military procurement.

A common misconception is that simulation perfectly represents reality. In practice, it's a continuous loop: real-world data is required to tune simulator parameters, and this validation must be repeated until the gap between simulation and reality is small enough to trust the results.

Long training cycles mean too many variables change between models, making it impossible to attribute improvements. By training smaller models end-to-end more frequently (e.g., in weeks, not months), research teams can better understand which specific "ingredients" led to better outcomes, enabling a more scientific process.

In military AI, being three months ahead in developing a frontier model is irrelevant if the organization is five years behind in adopting and integrating existing capabilities. The crucial race is not about invention but about effective implementation, an area where the US military's bureaucracy poses a significant disadvantage.

Current methods for updating models with new data suffer from catastrophic forgetting, forcing labs to periodically retrain models from scratch. This inability to continuously integrate new information into the same model is a fundamental technical and economic bottleneck that prevents the creation of a single, persistently learning AI.