# The Biggest Risk With AI Isn’t Job Loss. It’s Loss of Agency. #AI *Last Updated: February, 2026* History is comforting. Every major technological wave displaced some jobs, created others, and eventually expanded the economic pie. The loom didn't eliminate work. Electricity didn't. The internet didn't. So the instinctive move is to say, "AI will follow the same script." I catch myself reaching for that comfort too. And broadly, I agree with the pattern. But AI might represent a structural break that history doesn't fully prepare us for. Not because it automates faster. But because it automates differently. ## Why This Wave Feels Different Think about earlier waves of automation as replacing rungs on a ladder. Machines took over the lower rungs in the form of repetitive tasks, physical labour, basic calculation and humans climbed higher. From muscle to machine. From calculation to coordination. From execution to judgment. AI doesn't replace the lower rungs. For the first time, it operates directly in the domain of judgment, synthesis, and abstraction. > **It doesn't just execute instructions. It participates in shaping them.** ## The Two Shifts Worth Watching ### The Loop That Tightens Most historical automation was static. You built a machine. It performed a bounded task. Done. AI systems are different. They improve through feedback loops, observing outcomes, refining models, and closing the gap. Over time, the surface area requiring human intervention can shrink. Not from the bottom up, the way we're used to. But inward. Toward the core of decision-making. > **That doesn't mean humans disappear. It means the center of gravity shifts. And when the center shifts, leverage concentrates.** If you've ever watched a team gradually hand off more decisions to a well-tuned system, you know the feeling. At first, it's liberating. Then, quietly, you start wondering which parts of your judgment still matter. ### The Uneven Distribution of Adaptation Working effectively with AI requires meta-skills that are rare. You need to know where judgment truly matters. How to make tacit reasoning explicit. How to operate comfortably in probabilistic systems instead of deterministic ones. These aren't skills that diffuse automatically. We like to believe competence spreads evenly across populations. It rarely does. The printing press didn't make everyone an author. The internet didn't make everyone a publisher of consequence. > **So while employment may grow in aggregate, the distribution of agency may narrow.** The divide may not be between humans and machines. It may be between humans who can structure problems for AI and humans who cannot. I include myself in this uncertainty. On some days, I feel like I'm riding the wave. On others, I wonder if I'm just rearranging furniture on a shifting floor. ## The Commentary Trap Most commentary on this topic, on either side, extrapolates from general-purpose conversational interfaces. Chat windows. Prompting tricks. Coding copilots. That's surface-level AI. But here's where it gets interesting. Workflow-embedded, production-grade cognitive infrastructure is something else entirely. Think less "chat assistant." Think domain-native systems woven directly into legal workflows, financial analysis, medical diagnostics, and research pipelines. Something like Harvey in law isn't just a faster intern. It's a restructuring of how legal cognition gets distributed across a firm. > **Scale that pattern across domains over the next few years, and the question shifts from "Will AI replace jobs?" to "Who remains cognitively central?"** That's a different question. And a harder one. ## Anxiety vs. Redesign If more of us spent time redesigning our workflows instead of doom-scrolling AI headlines, this would likely be a net positive. But history offers a sobering reminder. There's very little precedent for large populations proactively re-architecting how they work in anticipation of structural shifts. We react. We adapt late. We optimize locally. That's just a human thing to do. History tells us jobs will change. What's less certain is how widely the capacity to remain cognitively central will be distributed. > **The risk isn't universal unemployment. It's uneven agency.** And that's subtler. Harder to detect. Harder to debate. But potentially more destabilizing. The real work ahead isn't resisting AI. It's learning how to think in partnership with it, deliberately and structurally. Because this time, the machine isn't just doing the work. It's increasingly shaping what the work is.