Algorithmic Management Doesn't Hurt Gig Workers Directly. It Hurts Their Autonomy, Which Hurts Everything Else

Based on the research ofAhmed M. Elsayed, Saleh M. Saleh, and Ibrahim G. Mahgoub, "How Algorithmic Management Shapes Agility in the Gig Economy: The Role of Job Autonomy," Sage Open, 2026

A new structural study of gig workers reframes a debate that has mostly been fought at the wrong altitude. Ahmed Elsayed, Saleh Saleh, and Ibrahim Mahgoub use a Job Demands-Resources model to test how five algorithmic management functions, monitoring, goal setting, scheduling, performance rating, and compensation, affect worker agility: the proactivity, adaptability, and resilience that let a gig worker respond well to a volatile job. Some of these functions do have direct negative effects on agility. But the headline result is sharper than "algorithmic management is bad for workers." All five functions show significant negative indirect effects on agility, and they travel through the same channel: they erode job autonomy first, and the loss of autonomy is what degrades agility. The damage is mediated, and the mediator is fixable.

The Diagnosis Changes the Prescription

This distinction matters because it points platform operators toward a different fix than the one they usually reach for. The common response to criticism of algorithmic management is to dial back monitoring, intensity, or scoring across the board, a blunt move that sacrifices the efficiency gains algorithmic coordination was built to capture. The paper's finding implies something more surgical: the problem is not that workers are watched, scheduled, or scored. The problem is that these functions, as typically implemented, strip workers of control over how, when, and under what terms they do the work. Restore the autonomy and you plausibly recover much of the agility loss, without necessarily giving up monitoring or rating as tools.

This is not a hypothetical design question. Two documented episodes in the ride-hailing industry show both halves of the mechanism: how algorithmic functions compress autonomy, and what happens when regulators or platforms try to hand some of it back.

The first concerns Uber and Lyft drivers in New York City, whose minimum per-trip pay is set by the city's Taxi and Limousine Commission using a utilization rate, the share of a driver's logged-in time that is spent on paid trips. Drivers and the New York Taxi Workers Alliance have argued that the platforms responded to this formula by locking drivers out of the app during slow periods, which inflates the utilization number and lets the companies hold pay down, while the companies have said the pay structure itself forces the restriction. Whatever the intent, the effect on the driver side is a scheduling function, nominally about matching supply to demand, that removes a driver's ability to decide when to log in and work. In 2025 the TLC revised the rules: lockouts now require 72 hours of advance notice and cannot exceed 16 hours, a direct, procedural restoration of a sliver of driver control over their own calendar, alongside a revised pay formula that, as Uber itself acknowledged, stopped automatically adjusting driver pay down each year based on the utilization rate alone.

The second episode is sharper, and less settled. In January 2026, the New York City Council overrode a mayoral veto to enact Local Law 52, requiring that high-volume for-hire services like Uber and Lyft show just cause, a bona fide economic reason, or a legal requirement, and give 14 days' notice, before deactivating a driver. The underlying complaint, voiced by the New York Taxi Workers Alliance and the bill's sponsors, was that drivers had effectively no autonomy over their own continued employment: a platform could end a livelihood unilaterally, with no process and no appeal standard. That is exactly the kind of autonomy erosion the paper's mediation model would predict degrades agility, since a worker who cannot predict or contest the terms of their own employment has little basis for the proactivity and resilience the researchers measure. But the law never took effect as written: Uber and Lyft sued in June 2026, and in late July a federal judge granted a preliminary injunction blocking enforcement, ruling the companies were likely to succeed in showing the notice requirement unconstitutionally impaired their driver contracts. The autonomy-restoring fix and the platforms' contractual counter-claim are now litigating the same question the paper poses analytically: how much say over the terms of work a platform has to hand back.

Algorithmic management's damage to worker agility is not mainly direct. It runs through eroded autonomy, and autonomy is the part a platform can choose to redesign.

What the Research Record Already Showed

The paper's finding is not an isolated claim; it extends a decade of research on exactly this mechanism in ride-hailing specifically. Min Kyung Lee, Daniel Kusbit, Evan Metsky, and Laura Dabbish's 2015 CHI paper, "Working with Machines: The Impact of Algorithmic and Data-Driven Management on Human Workers," interviewed Uber and Lyft drivers and coined the term "algorithmic management" to describe how the apps assign work, provide informational support, and evaluate performance, documenting that drivers actively tried to make sense of and respond to these algorithmic features. A year later, Alex Rosenblat and Luke Stark's "Algorithmic Labor and Information Asymmetries," a nine-month study published in the International Journal of Communication, found that Uber exercised significant indirect control over how drivers worked and that drivers risked deactivation if their acceptance rate fell too low, a direct, pre-reform illustration of a scheduling-adjacent metric functioning as an autonomy constraint rather than a quality signal. Lisa Möhlmann and Lior Zalmanson's 2017 study, presented at the International Conference on Information Systems and titled "Hands on the Wheel: Navigating Algorithmic Management and Uber Drivers' Autonomy," went further and showed drivers actively resisting, gaming, and working around the algorithm to reclaim some of the control the platform had taken from them, exactly the proactive, adaptive behavior the new paper's agility construct is trying to capture, and exactly what withers when autonomy keeps shrinking.

Read together, these studies and the new structural paper tell a consistent story across more than a decade. Workers do not passively absorb algorithmic constraint. They adapt to it, resist it, or are worn down by it, and the degree to which they retain any say over the terms of their own work is the variable that predicts which of those three outcomes occurs. That is a testable, design-relevant claim, not a moral one, and it is why the mediation finding matters more than another dataset confirming that gig work is stressful.

The Part That Is Actually Negotiable

None of this argues that platforms should abandon monitoring, scheduling algorithms, or performance ratings. Those functions exist because they solve real coordination problems at a scale no human dispatcher could manage, and the paper does not claim otherwise. What it establishes is that the harm these functions cause to worker agility, and by extension to the proactive, resilient behavior that keeps a platform's service quality up, travels overwhelmingly through one variable that platforms control the design of: how much autonomy survives the implementation. A monitoring system that reports on a worker without stripping their discretion over how to meet the goal is not the same intervention as one that reports on a worker while also dictating the method, even if the two look identical on a dashboard.

The New York cases show that this redesign does not have to wait for voluntary platform reform, even if it also does not arrive uncontested. Regulators intervened on exactly the two functions, scheduling access and deactivation, where autonomy loss was most visible and most consequential, and in both cases the intended fix was procedural rather than punitive: notice periods, caps on lockout duration, and a stated-cause standard before an irreversible decision. The lockout rule survived; the deactivation notice rule is enjoined pending litigation. Neither change touches the underlying algorithmic coordination that makes the platforms efficient. Both touch only the layer where a worker's say over their own work day either exists or does not, which is exactly the layer a court, not just a regulator, now has to rule on.

The practical implication for anyone designing or regulating algorithmic management is to stop asking "how do we make this less intensive" and start asking "how much control does the worker retain under this specific function, and can we give more of it back without giving up the coordination benefit." The paper's mediation model suggests that question, asked function by function, is where the actual leverage over agility and performance sits, not in the blunt instrument of dialing surveillance up or down.