The Algorithm Is Your Manager Now — And It Can't Be Fired

Imagine a supervisor who assigns your shifts, tracks your location in real time, scores your performance after every task, adjusts your pay rate without notice, and can end your livelihood with a single automated decision — all without ever meeting you, hearing your side, or being required to give a reason. Now imagine that no employment law in your country recognizes this supervisor as a supervisor at all. That is not a hypothetical edge case. It is the daily legal reality for tens of millions of gig workers worldwide.

In June 2025, the International Labour Organization convened a formal debate on algorithmic management in platform work^[3] — an acknowledgment, unusually blunt for the ILO, that the legal frameworks governing labor relations were built for a world where bosses were human. Around the same time, Equidem, the labor rights organization, released a global report^[4] documenting the working conditions of gig and platform workers across multiple continents. The picture that emerged was consistent: the tools of management — task assignment, performance evaluation, pay-setting, discipline, termination — had been automated. But the obligations of management had been quietly left behind.

The mechanism is not complicated, which is part of why it has been so effective. Most platform companies classify their workers as independent contractors. Contractors, legally speaking, work for themselves. They are not employees, so the company does not owe them minimum wages, overtime protections, sick leave, collective bargaining rights, or procedural fairness before dismissal. That classification made some intuitive sense in an older model of contracting, where a plumber or a freelance writer genuinely set their own terms, negotiated their own rates, and worked across multiple clients without any single one controlling their day. None of that describes a driver whose every move is monitored by an app, whose surge pricing is set by a remote server, and whose rating can drop below a threshold — quietly, automatically — and simply stop sending them work.

What the platforms built, structurally, is a management layer. What they called it, legally, is software. The distinction is doing enormous work — for them.

The Architecture of Control Without Accountability

Algorithmic management is not a new phenomenon in 2025, but the Equidem report placed it in sharper relief by documenting its reach across sectors that don't always make headlines. Delivery riders in South and Southeast Asia. Domestic cleaners booked through apps in the Gulf. Care workers matched to clients by platforms in Europe. The industries vary. The underlying structure does not. In each case, a platform mediates the relationship between worker and customer while claiming to be a neutral marketplace — not an employer — and the algorithm that governs the worker's day is framed as a technical tool, not a managerial one.

The specific mechanisms of algorithmic control tend to cluster around a few functions. Assignment systems determine which workers receive which jobs, based on proximity, historical ratings, acceptance rate, and parameters the worker cannot see or contest. Pricing systems set what workers earn per task, adjusting dynamically in ways that can increase income in high-demand periods but also compress it during platform promotional campaigns or in response to competitor pricing. Evaluation systems aggregate customer ratings and other behavioral signals — time to acceptance, completion rate, reported incidents — into composite scores that determine a worker's future access to work. And deactivation systems enforce the platform's standards by suspending or permanently removing accounts, typically with minimal notice and no appeal process that resembles anything courts would recognize as due process.

“The algorithm does everything a manager does — except carry the legal responsibilities that management has always implied.”

Each of these functions, in a traditional employment setting, would carry legal weight. A supervisor who sets your pay, assigns your tasks, evaluates your performance, and terminates you is exercising employer authority. Courts and labor regulators have built bodies of law around exactly that authority — protections against arbitrary dismissal, requirements for transparency in performance assessment, rights to contest adverse decisions. When those same functions are performed by software, the platform argues that no such relationship exists. The worker is self-employed. The algorithm is just a tool. The company is a technology intermediary.

A Legal Fiction Built to Last

What makes the contractor classification so durable, despite mounting pressure from regulators in the UK, EU, California, and elsewhere, is that it is not simply a lie. It is a structure that was deliberately engineered to look like independence. Workers do set their own hours, in the sense that they can log on when they choose. They do work across multiple platforms, at least in principle. They do bear their own costs — vehicle maintenance, phone data, equipment — which looks formally like the risk-bearing of a business owner. The platforms have invested heavily in preserving these features, because the features are not just branding. They are legal architecture.

The UK Supreme Court's 2021 ruling that Uber drivers were workers rather than independent contractors^[2] — a case that took years to reach that outcome — showed that courts could see through the structure when pushed. The EU's Platform Work Directive, finalized in 2024^[1] after years of negotiation and significant lobbying pressure from platform companies, established a rebuttable presumption of employment for platform workers, shifting the burden of proof onto companies to demonstrate that a relationship is genuinely one of self-employment. These are real shifts. But they are also geographically limited, slow to implement, and perpetually outpaced by the speed at which platforms can restructure their operational models to preserve the legal outcome they need.

The ILO debate in June 2025 surfaced a deeper problem: there is no global floor. A platform operating across twenty countries can apply different classifications, different pay structures, and different terms in each jurisdiction, calibrating its legal exposure market by market. In countries with weak enforcement, the contractor model is essentially unchallenged. In countries with stronger protections, platforms reclassify workers, adjust the interface, and continue. The algorithm does not care which country it is running in. The legal vulnerability of the workers it manages does.

What the Score Doesn't Tell You

Beyond the classification question, algorithmic management introduces a specific kind of opacity that traditional employment law was not built to handle. When a human manager fires you, there is at least a record — an HR process, a written warning, a meeting. The decision passed through a person who can be questioned, who made a judgment call, who can be held accountable in some traceable way. When an algorithm deactivates your account, the decision may have been made by a model trained on aggregate behavioral data, running automated fraud detection, or responding to a spike in customer complaints that you were never told about and cannot inspect.

“A customer's one-star rating, logged in thirty seconds, can trigger a cascade that ends a worker's income — with no human ever reviewing what actually happened.”

The Equidem report documented cases where workers were suspended or deactivated following customer reports that they could not access or contest, where rating thresholds were changed without notice, and where the appeals process — where one existed at all — amounted to sending an email and receiving an automated response. This is not a bug in an otherwise functional system. The opacity is structurally useful. A system that cannot be inspected cannot easily be litigated. A decision that was made by a model rather than a person is harder to pin on anyone in particular.

There is a labor-economics concept that captures part of what is happening here: monopsony, a market condition where one buyer — in this case, a platform that controls access to a large pool of customers — has significant pricing power over sellers of labor. The platform does not need to be the only option available to a worker for this dynamic to operate. It only needs to be large enough that leaving it imposes real costs, and structured in a way that makes the worker's investment — in ratings, in reputation, in familiarity with the interface — non-transferable. The rating a driver built on one platform does not move with them. The history vanishes. Starting over is expensive.

The Productivity Argument and Its Limits

The standard platform response to this framing is that flexibility is real and that many workers genuinely value it. This is not entirely false. Survey data on gig workers consistently shows a segment — typically smaller than platforms imply in their lobbying materials, but real — that uses platform work as genuine supplemental income, prefers variable hours, or is combining it with other work or caregiving in ways that a traditional employment contract would make harder. Collapsing all gig workers into a single category of exploitation misses the actual diversity of the workforce.

But the flexibility argument has always contained a category error. The question is not whether flexible work should exist. It is whether flexibility should require forfeiting all the protections that labor law spent a century building — minimum wage guarantees, injury compensation, the right to organize, protection against arbitrary termination. These are not things that workers traded for flexibility. They are things that platforms structured the relationship to avoid, and then marketed the absence of as a feature. The Equidem report was particularly pointed on this: in many of the markets it studied, the workers most dependent on platform income — those for whom it was not supplemental but primary — were also the least likely to benefit from the flexibility framing, and the most exposed to the risks the classification was designed to exclude.

“Flexibility became the price workers paid for protections they were never offered a chance to keep.”

Who Is Actually Accountable When the System Fails

When a gig worker is injured on the job, there is typically no workers' compensation system to absorb the cost. When a worker is deactivated on the basis of a fraudulent complaint, there is typically no grievance mechanism with any binding force. When pay rates are cut overnight, there is no obligation to negotiate or even to notify. The platform, by virtue of the contractor classification, has offloaded every material risk onto the worker while retaining every lever of control. That is not a market outcome. It is a legal architecture — one that requires active maintenance by lobbyists, lawyers, and occasionally legislators.

What the ILO debate signaled, and what the Equidem report made empirically concrete, is that this architecture is beginning to face coordinated scrutiny on multiple fronts simultaneously. The EU's Platform Work Directive is one pressure point. Litigation in the UK, Australia, France, and several US states is another. Organizing among gig workers — historically difficult given the dispersed, high-turnover nature of the workforce — has produced results in places like Denmark and Spain that would have seemed unlikely a decade ago. And the growing body of research on algorithmic management is making it harder for platforms to claim that their systems are neutral technical tools rather than substitutes for human supervision.

None of this means the contractor model is collapsing. Platforms have shown a consistent ability to adapt their structures to preserve the legal outcome they need, and the global patchwork of labor law gives them enough room to maneuver. But the friction is increasing. Courts and regulators are developing new vocabularies for what algorithmic management actually is, and a few jurisdictions are starting to write those vocabularies into law. The question is whether the law can develop fast enough — and with enough cross-border coordination — to catch up with a system that was always designed to stay one step ahead of it.

The deeper issue is one of institutional honesty. What algorithmic management has done, at scale, is disaggregate the concept of the employer — taking the control while leaving the obligation — and distributed that disaggregation across enough software layers that responsibility becomes genuinely hard to locate. That is not the neutral result of technological progress. It is a choice, encoded into a business model, that shifts costs from the company to the worker and calls the shift efficiency. Naming that choice accurately is where any serious regulatory effort has to begin.

References

1. Directive - EU - 2024/2831 - EN - EUR-Lex (eur-lex.europa.eu)
   Created a rebuttable presumption of employment for platform workers, shifting burden of proof onto companies to demonstrate self-employment.
2. United Kingdom Uber Drivers Are Workers Entitled To Employment Law Protections Supreme Court Rules (loc.gov)
   Established the 2021 legal precedent that Uber drivers qualify as workers entitled to employment protections, not independent contractors.
3. Algorithmic management in the workplace (ilo.org)
   Defines algorithmic management as systems that organize, assign, monitor, and evaluate work while reducing worker-manager contact.
4. Realising Decent Work in the Platform Economy: Addressing Intermediated Platform Employment in Food Delivery and Data Work – Equidem (equidem.org)
   Documents systemic abuse of food delivery riders and data workers across multiple continents, revealing wage theft and unsafe conditions.