Soldier vs. Algorithm: Where the US Army Draws the Competence Line in an AI-First Future

A provocative new analysis from the Modern War Institute at West Point is forcing military planners to confront a deceptively simple question: In an Army that increasingly relies on artificial intelligence, what does a soldier still need to know?

Published on September 11, 2026, by researcher Dorothy M. Reid, the article “What Must a Soldier Still Know? Drawing the Competence Line in an AI-First Army” dissects the human-machine boundary that will define future battlefields. Rather than celebrating AI as a cure-all, Reid’s piece challenges commanders to decide which core competencies must remain irreducibly human—and what happens when they forget that line.

Why a ‘Competence Line’ Matters

The central argument revolves around the risk of competence atrophy. Reid frames AI not as a replacement for soldiers but as a force multiplier that can degrade the very skills it is meant to augment. When algorithms take over tasks such as terrain analysis, logistics planning or even early tactical recommendations, the soldier’s ability to perform those functions independently may erode—leaving units vulnerable if AI systems are jammed, spoofed or simply wrong.

“The concern is that overreliance on AI will produce soldiers who can execute when the machine works perfectly but collapse when it doesn’t,” the article warns, according to the MWI piece. This echoes long-standing military principles of mission command and battlefield resilience, now refracted through the lens of machine learning.

Which Skills Stay Human?

The debate zeroes in on a handful of non-negotiable human competencies. Judgment—especially moral and ethical judgment in lethal situations—tops the list. Reid argues that no algorithm should substitute for a commander’s understanding of mission intent, proportionality and the law of armed conflict. Basic tactical reasoning is another: a squad leader still needs to read terrain, assess risk and improvise without waiting for an AI recommendation.

Also highlighted is the ability to verify AI outputs. The piece points to well-documented problems of machine bias, data poisoning and cybersecurity threats. A soldier who cannot sense-check an artificial intelligence’s conclusion becomes a dangerous link in the kill chain. Consequently, the competence line must protect the soldier’s role as a “human in the loop”—not a rubber stamp.

Modernization Meets Human-Machine Teaming

The analysis lands amid a sweeping U.S. Army modernization effort that places AI-enabled systems at the core of its future force design. Ground vehicles, aviation platforms and network tools are increasingly built around human-machine teaming concepts. The Pentagon’s Responsible AI strategy explicitly demands that humans remain responsible for decisions involving the use of force. Yet the practical question—how much understanding the human must bring to that table—remains fuzzy.

Reid’s piece pushes doctrine writers to go beyond platitudes. If an AI can predict maintenance failures, map an enemy’s supply routes and generate a recommended course of action within seconds, does a young officer need to master those skills from scratch? The answer, the article suggests, is a qualified yes—because the circumstances that break the machine are precisely the moments when foundational training matters most.

The Risk Management Equation

Underpinning the entire discussion is a calculus of risk. Reid frames the competence line as a deliberate trade-off: the efficiency gains from AI delegation versus the survivability loss when soldiers are stripped of deep understanding. Errors become catastrophic when confidence in a tool outpaces the ability to spot its failure. The article invokes examples from early autonomous driving and aviation, where overautomation led to skill fade and high-profile accidents.

Cybersecurity adds another dimension. A battlefield AI fed bad data could betray its unit, yet a soldier who lacks independent analytical muscle may not detect the ruse. The piece stops short of prescribing a one-size-fits-all rule, but insists that every capability handed to an algorithm must come with a training plan that preserves fallback human competence.

Training for an Unpredictable World

One of the most practical takeaways is a call to reshape how the Army trains. Instead of measuring only performance with AI tools, exercises should test soldiers in deliberately degraded modes—no network, corrupted data, scripted AI failures. Such “denial drills” would expose whether the competence line has been drawn in the right place. If pilots still practice hand-flying under failed systems, foot soldiers must do the same with the cognitive tools of their trade.

The analysis also nudges the Army to define what “AI-first” really means. Is it a philosophy that humans are always in control and merely assisted, or a design principle that assumes machines lead until a human veto? Clarifying that posture is essential before drawing any competence line that will stick.

A Conversation Rather Than an Answer

Reid does not hand down definitive rules, and that is precisely the point of the MWI piece. The competence line is not a fixed standard but a live-wire debate that will shift as technology matures. The article’s real value is in framing the questions that institutional inertia might otherwise skip: What must every soldier, regardless of rank, still know? How much is “enough” when the machine knows more?

In an era when the Army is racing to integrate AI from the tactical edge to the strategic rear, the risk of losing day-one soldiering skills has never been sharper. Reid’s intervention serves as a crisp reminder that modernization is not just a hardware problem—it is an intensely human challenge of drawing the right lines in the sand before a crisis draws them for us.