Why most defense AI pilots will fail and what it will take to fix them

Most defense AI pilots will fail. It’s not because the models are inadequate, but because they can’t access the right data in time. That’s the uncomfortable reality emerging from a growing body of reporting across defense and technology outlets:

• • Breaking Defense recently published this story on how the Army’s Scarlet Dragon exercise is adding commercial data to battlefield data flows.
    
    Why this matters: The Scarlet Dragon exercise is explicitly testing how commercial data can be ingested, routed, and used alongside military data in real time. The reason is straightforward: in many cases, commercial data is simply faster.

  • Business Insider published a new report discussing how Army soldiers are drowning in battlefield data and that AI is needed to make sense of it all. A complementary piece from Wired explores how the Army is attempting to solve the problem through an AI-enabled system that aggregates hundreds of disparate data sources into usable insights.
    
    Why this matters: The growing reality is that U.S. forces are now overwhelmed by the sheer volume of available data, including during large-scale exercises like Dynamic Front, where AI systems are tested against real-world data flows. For soldiers, more inputs no longer translate into better decisions.

  • Stars & Stripes in April published this piece about how the Army is working to turn data overload into battlefield advantage.
    
    Why this matters: The Army is standing up a dedicated data operations capability to actively manage and route information across the force – recognizing that data does not move itself and needs to be at the point of use.

  My takeaways

  • AI is not the advantage. It is the bridge between data and decision.
    Without AI, data remains siloed, people become the bottlenecks, and decision timelines expand under pressure. With AI, disparate data becomes connected, context emerges across domains, and decision timelines compress. But even here, there is a risk. AI systems are only as effective as the data they can access. If pipelines are constrained, delayed, or incomplete, the model does not fail loudly. It simply produces less useful answers.
  • The true decisive advantage is having the RIGHT data at the RIGHT time.
    AI systems like the ones described above are only as effective as the data they can access. Commercial data offers a path forward, with a vast selection of high-quality, mission-relevant information available almost instantaneously. The question is no longer whether to use commercial data, but how to integrate it effectively.
  • Data has to move to matter.
    The challenge for the Army is ensuring that relevant data, often commercial, can be discovered, integrated, and delivered fast enough to matter.Pipelines break, interfaces fail, classification boundaries slow everything down, and in contested environments, even well-designed systems degrade. Architecture like pipelines, APIs, and cross-domain solutions is what determines whether commercial data can be ingested at operational speed, whether AI systems receive timely inputs, and whether insights reach the edge before they are irrelevant.Put differently, getting data will soon no longer be the bottleneck. Routing the data to the point of decision will be. And routing is not a static design problem. It is an operational discipline, one that must be continuously tested, adapted, and hardened under real conditions.

  Taken together, these shifts point to a deeper change in how real advantage is created:

  • Securing fast access to high-quality commercial data; reducing reliance on slower, more expensive, or internally generated data
  • AI that effectively bridges the gaps between silos of disparate data
  • Getting data to the tactical edge at mission speed – not just gathering it in the cloud.

  A decision made in 30 seconds versus 30 minutes is the difference between mission success and mission failure. Between action and missed opportunity. Success will only come when defense programs can turn data into decisions, reliably, under pressure, and in time to matter.