At Amazon's Manesar warehouse in India, a system called ADAPT continuously measures worker productivity, with penalties triggered by missed targets or mistakes detected on CCTV. The image of a worker pacing a warehouse aisle under camera-enforced quotas is not new. What is new is that the same technical stack — pace tracking, quota assignment, automated discipline — is being ported to the keyboard.
The cyberpunk image of "low life" is not a soup line. It is a person working ten hours under fluorescent lights with a scanner that beeps every fourteen seconds and a screen that says their pace is below quota. Cory Doctorow put the trajectory plainly: the future of Amazon coders is the present of Amazon warehouse workers.
The warehouse is where the playbook was perfected
Algorithmic management did not start at Amazon, but the modern template was refined inside its fulfillment centers. Instructions arrive through workstation displays and barcode scanners. Each scan triggers the next task. Every movement is logged. A pace target is computed for every worker against every product category, recalibrated continuously as the worker speeds up.
The worker is not managed by a human. The worker is managed by a model that sets the next quarter's quota from data the camera collected last quarter. A floor supervisor exists, but the supervisor's role is to enforce numbers the algorithm produced, not to negotiate them. There is no manager to talk to about an unfair shift, because the rule was set in code, and the code is owned by a company whose lawyers know it does not have to explain itself.
This is the structural innovation. The traditional foreman could be persuaded, bargained with, or worked around. The algorithm cannot. It is the same management function with the human discretion stripped out, which means the points of leverage workers historically used to push back have been removed by design.
The same toolkit does anti-union work
A 2025 Northwestern study published in Sage's Socius journal, Weaponizing the Workplace, documented something colder than productivity tracking. During the union campaign at Amazon's Bessemer, Alabama warehouse, the company repurposed the same algorithmic management tools used to set quotas to gauge union sympathies and target anti-union messaging.
Read that sentence again. The system that tells the worker what to do also tells management how to break the worker's organizing. The data flowing into the productivity model is the same data flowing into the union-suppression model. There is no firewall, because there is no architectural reason for one to exist.
This is the part of the story that gets least discussed because it does not photograph well. Algorithmic management is not just a productivity question. It is a labor-relations question. The tools that optimize a fulfillment center can be re-aimed at any collective action that threatens the firm's interests. The Bessemer record demonstrates the re-aiming has already happened.
Now port the stack up the org chart
The PR for "AI-powered productivity tools" aimed at white-collar work is, technically, the same surveillance and pace-enforcement layer applied to a keyboard instead of a scanner. Pull-request velocity dashboards. Per-keystroke productivity scores. Auto-generated performance reviews from Slack and email content.
Each is shipping today in mainstream productivity suites. They default to manager-visible reporting. The dashboard the worker sees is a subset of the dashboard their VP sees. The asymmetry is the product.
A 2025 arXiv working paper, Fulfillment of the Work Games, describes how measured workers rationally adapt by gaming the metrics — burning out on metric-favorable tasks, hiding work the system cannot count, atrophying skills the model does not score. The same dynamics will appear in software, sales, and consulting as those teams get instrumented to the same depth.
Why this is the labor story of the decade
The dominant narrative of AI and work is replacement: the model does the job, the worker is laid off. The quieter, more durable story is management: the model does not do your job, but it decides whether you keep it, how fast you do it, and whether your organizing counts as misconduct.
That story has the advantage, from capital's perspective, of being deployable today, with off-the-shelf tools, against any cognitive worker on a corporate network. Replacement is constrained by what the model can do. Management is constrained only by what the firm will install. The constraint is collapsing.
The forward question for white-collar workers is not whether AI will take their jobs. It is whether the scanner that beeps at the warehouse worker every fourteen seconds will, by 2028, be the same scanner counting their commits.