The Involuntary Pilot Program

The GSA just lost 40% of its workforce and is now automating a million work hours to compensate. This isn’t a cautionary tale. It’s a preview of the substitution-versus-augmentation split playing out everywhere, and of what the 4I model actually protects against.

Read that again slowly. Forty percent of the people. Gone. Now AI fills the gap.

Nobody at the GSA called this a pilot program. But that is exactly what it is. A large institution, stripped of human capacity by circumstance rather than choice, is now discovering in real time which of its functions an AI system can absorb. The results will be instructive for every organization watching from the sidelines.

What the GSA is running involuntarily, the rest of the economy is running voluntarily. And the data from this week suggests the experiment is accelerating.

Reshape Versus Replace: A Distinction That Matters

A new BCG analysis published this week projects that 50 to 55 percent of U.S. jobs will be substantially reshaped by AI within the next two to three years. The headline finding is framed reassuringly: AI will reshape more jobs than it replaces. Most roles will remain.

But the fine print deserves attention. BCG draws a clear line between substitution and augmentation. Substitution is when AI takes over the core function of a role entirely. Augmentation is when AI handles portions of a role while the human handles the rest, often at higher levels of complexity. For call center representatives and routine software engineers, the BCG analysis sees substitution already happening at scale via agentic AI. For more complex technical and leadership roles, it sees augmentation.

The critical variable is not how sophisticated your current work feels. It is how much of your work depends on physical presence, sustained human interaction, or judgment that cannot be cleanly separated from context and relationship. Those are the functions that resist substitution. The ones that don’t resist it are already being automated.

This is not a comfortable framework for people who assumed their white-collar credentials insulated them. A Gallup survey released this week found that 18 percent of all U.S. employees now say it is very or somewhat likely their job will be eliminated within the next five years due to AI or automation. That number rises sharply among workers in office and administrative support, finance, and legal roles. The workers closest to the substitution side of BCG’s model are the ones already feeling it.

The Jobs That Are Most Exposed

Inside Higher Ed reported this month on the accumulating research from Yale, Stanford, and MIT identifying which career fields face the highest exposure to AI-driven automation or augmentation. The consistent finding across all three: jobs related to writing and coding rank highest.

This is a remarkable data point for universities to sit with. The degrees that were supposed to be AI-proof because they required “complex reasoning” and “creative output” are now among the most exposed. The research and writing skills that business schools, law schools, and journalism programs have built their reputations on are precisely the skills that large language models handle most fluently.

What does that mean for students? Job-market experts quoted in the Inside Higher Ed piece are direct: colleges cannot afford to ignore these projections. But the piece also notes what those projections do not show. Exposure is not the same as elimination. The most exposed jobs are not necessarily the ones that disappear. They are the ones that get hollowed out at the routine level, while demand rises for the functions that AI cannot provide.

That distinction is the entire game.

What the GSA Experiment Actually Tests

Return to the GSA for a moment. When an institution loses 40 percent of its workforce and responds by automating a million work hours, it is not automating the same work those 40 percent were doing. It cannot. AI systems do not replicate human judgment, institutional knowledge, or relationship-based coordination. They automate the processable. The forms, the routing, the retrieval, the classification, the drafting of standard documents.

What remains after a million automated work hours? The decisions about what to automate and how. The verification of what comes out. The communication of changes to stakeholders inside and outside the agency. The management of the humans who are left, in an environment that is now smaller and more uncertain. The navigation of whatever the automation gets wrong, which it will, and the accountability for fixing it.

None of that is automatable. All of it is hard. And all of it maps directly onto the four domains that define professional value in an AI-saturated environment.

Interpretation is what the remaining GSA workforce has to do every time an AI system produces an output that someone has to decide to act on. The model can draft the memo. A human has to read it, judge whether it captures the actual situation, and own what goes out under the agency’s name. That interpretive layer does not get cheaper when AI writes the first draft. It gets more important, because the volume of drafts increases while the human capacity to review them stays flat.

Integration is the leadership work of deciding which functions to automate, in what sequence, with what guardrails, and how to redesign the workflows that exist around those functions. This is not a technical problem. It is a strategic and organizational one. The GSA’s IT leadership is doing this work right now. No AI system is doing it for them.

Interpersonal connection is what keeps an organization coherent when 40 percent of its people have left and the remaining 60 percent are watching automation absorb tasks that used to belong to their colleagues. The trust, communication, and culture work required in that environment is not a soft skill. It is a core operational function. Institutions that get it wrong will see the remaining talent leave. The ones that get it right will stabilize faster.

Imagination is what it takes to ask the right question at the beginning: not “how do we automate what we used to do?” but “given what AI can now handle, what should we be doing that we couldn’t before?” The GSA’s situation is extreme, but every organization will face a version of this question. The professionals who can reimagine scope and purpose in an AI-enabled environment are the ones who create value that compounds. The ones who just prompt the tools are doing the equivalent of operating the copy machine.

The Involuntary Experiment Goes Voluntary

The GSA did not choose this situation. Workforce cuts came first, automation followed as a response. But the experiment the GSA is running under duress, corporate America is running by design. The Guardian reported last week that more than 165,000 tech workers were laid off in the past year, with Microsoft, Amazon, Block, Oracle, and Meta all making major cuts explicitly tied to AI-driven operational efficiency.

The pattern is the same whether it is forced or chosen. Headcount falls. Automated work hours rise. What remains is the judgment layer.

BCG’s report says workforce strategy “cannot sit downstream of automation.” That is the most important sentence in any business publication this week. Organizations that treat AI as a tool to hand to existing workers are already behind the ones treating it as a reason to rethink what work is and who does it. The leaders making that rethinking happen are not the most technically sophisticated people in the building. They are the most strategically clear, the most interpersonally skilled, and the most willing to imagine something different.

The GSA’s involuntary pilot program will eventually produce data. Which functions automated cleanly. Which ones failed. What the gaps cost. Other agencies, companies, and universities will learn from it.

The question is what you are doing right now to make sure you end up on the right side of that substitution-versus-augmentation line. Not just AI-literate. Not just prompt-capable. Actually skilled at the work that AI can hand off to you, but cannot do itself: reading the output with judgment, integrating it into decisions, holding the relationships together, and imagining what comes next.

Those four things are not the things the model does. They are the things you do. And they are the only things that are getting more valuable.

Andrew Perkins is the author of After the Grind: Rethinking Your Business Career in the Age of Artificial Intelligence and Robotics and Chair of the Department of Marketing and International Business at Washington State University’s Carson College of Business.