The fact that governments think alike means that larger governments are less able to deal with problems that don’t fit the model.
Jonathan Aberman
Washington is having the wrong conversation about artificial intelligence.
Federal discussions have focused almost exclusively on safety guardrails, procurement rules, and oversight frameworks. These concerns are real, and I do not intend to ignore them. But they dodge a more subtle threat that I think deserves more attention. What would happen if AI quietly trained government workers to think the same way?
This is already happening. You can tell by looking at how work is starting to look and sound across agencies.
As generative AI becomes part of an agency’s daily workflows (taking memos, integrating policy options, preparing briefings), output will naturally gravitate toward statistical centers. It’s not a flaw. That’s how the technology was built. AI learns from vast datasets and optimizes output to reflect existing patterns. The result is proposals that are cognitively convergent, well-structured, defensible, and largely indistinguishable from one another. Yes, it’s efficient. But in a government faced daily with novel, complex, high-stakes decisions, efficiency built on sameness can intangibly multiply risks.
Consider what the federal government is doing. Decisions are often made in situations where there is no clear precedent. Examples include pandemic response, cybersecurity crises, supply chain disruptions, and geopolitical shocks. The value of human judgment in moments like these is not just procedural. It’s the ability to see what the model doesn’t know, ask questions the dataset didn’t consider, and keep two opposing ideas in mind long enough to find a third way. Cognitive diversity suffers when teams from different agencies all rely on the same AI-generated analyzes and frameworks.
In my years of working at the intersection of business transformation, government, and academia, I’ve seen organizations misinterpret process improvement as strategic progress. AI deployment in the public sector is at significant risk of following that pattern. Leaders measure success by adoption metrics such as number of tools deployed and number of automated workflows. However, actual tests are difficult to quantify. Has AI made people think better or worse?
This distinction is very important in governments, where policy decisions can have consequences that worsen over time. Recommendations that seem reasonable because they match what the AI has surfaced may be precisely the kinds of recommendations that fail catastrophically in cases that determine the credibility of government agencies and public trust.
From a workforce perspective, this is not less urgent, but more urgent. Government agencies absorbed significant layoffs. Remaining employees are being asked to do more, faster, leveraging AI as a productivity bridge. In that environment, respecting the AI’s output is the path of least resistance. To embrace the AI framework, adopt the language of AI and move on to the next task. That’s understandable. It also happens when organizations silently lose the independent judgment that makes policies effective. And that is where we fall prey to the biggest fallacy of the AI era: that efficiency is the only metric that matters.
So what should leaders do?
First, recognize that using AI is not the same as understanding AI. Make sure all government employees understand that having the tools is just the starting point. In fact, when used properly, GenAI can amplify human judgment and independent thinking. Humans are instinctively wired to create and value novelty. We will use all the tools, including AI, to push us in that direction. But AI is a very unique tool. Your feedback loop with your users requires more than a “how-to guide.” That requires training to a different standard. This means recognizing the benefits of AI for the task and the unique ability of humans to use AI in their current roles.
Second, build an AI adoption strategy that is role-specific rather than blanket. Not all functions of government have the same uniqueness. Automating repetitive administrative tasks is easy and valuable. However, roles that require integration, judgment, discretion, or public accountability are different. To treat them the same is to inadvertently automate the very functions that make government institutions legitimate.
Third, we track cognitive performance over time, not just tool adoption. The question agencies should be asking is not “How many employees are using AI?” The question is, “Are our people developing better judgment or are they becoming more dependent?” These do not follow the same trajectory and have very different outcomes for employee outcomes. Leaders who cannot honestly answer that question are acting blindly.
The federal government is currently making a transformative bet on AI. Some of those bets may pay off. But speed and efficiency are not the only measures of success. The fact that governments think alike means that larger governments are less able to deal with problems that don’t fit the model.
The age of AI will not diminish the need for natural human thinking in public services. That makes them more valuable and more fragile than ever before. Washington should start governing accordingly.
Jonathan Aberman is co-founder and CEO of Hupside, partner at Ruxton Ventures, and founding dean of Marymount University’s School of Business, Innovation, Leadership, and Technology.
Copyright © 2026 Federal News Network. Unauthorized reproduction is prohibited. This website is not directed to users within the European Economic Area.
