Algorithmic management has moved beyond gig work
Algorithmic management still brings to mind a courier waiting for an app to assign the next delivery. That image is now too narrow. The same managerial logic has entered ordinary offices, factories and other workplaces. Software can allocate tasks, set pace, evaluate performance and turn monitoring data into decisions.
New European research gives a more precise account than “workplace AI is good” or “workplace AI is bad.” Design matters. Different controls can pile up. Tracking a person's body is not the same as recording activity in a business application.
The research arrives during a live policy debate. The European Commission's second-stage consultation on a Quality Jobs Act runs until 28 September 2026. Algorithmic management, workplace AI and protection from excessive monitoring are among its five priority areas.
This is not only about app workers
The Joint Research Centre's AIM-WORK survey collected responses from 70,316 people aged 16 to 65 in all 27 EU member states. Fieldwork ran from October 2024 to January 2025. Its methodology report describes a mobile telephone survey using random-digit sampling, national samples and weighting designed to make the results representative.
A JRC report on the platformisation of work says more than 90 percent of EU workers use digital devices and about one third report using AI for work. Digital monitoring is common, especially for working hours and entry or exit. Algorithmic management is less prevalent, but already includes automated task allocation and performance evaluation.
The word "platform" can therefore mislead. A workplace does not need to resemble a delivery marketplace to adopt platform-style control. Someone may have a regular contract and a human supervisor while a scheduling system sets shifts, a scanner determines pace, or a dashboard ranks output. The manager has not vanished. Parts of management have been embedded in software.
Measurement and direction are different powers
The latest JRC analysis of algorithmic management and working conditions separates individual practices instead of putting every workplace technology in one basket. Across the survey, algorithmic management was associated with lower autonomy, less ability to take breaks and more work-related stress. The size and direction of those relationships varied by technology.
Systems that directly regulated task execution and pace had the strongest associations with lower worker discretion and greater work intensification. Exposure to several practices also tended to reinforce negative outcomes. These are associations, not proof that software caused every result. The authors also found substantial differences between countries and organisational settings.
That qualification matters. A scheduling tool that respects employee preferences is not equivalent to one that changes shifts without an appeal route. A quality-control alert may catch a dangerous fault, while an opaque score may penalise workers for conditions outside their control. Labels such as "AI-powered" or "automated" reveal less than the actual decision path.
Watching a screen is not watching a body
A separate JRC paper on digital monitoring examines ten forms of oversight and draws a particularly useful line. Digital activity monitoring had modest negative or neutral associations with job quality. Physical tracking through CCTV, wearables and GPS was consistently associated with poorer conditions, including reduced autonomy, work intensification, limited flexitime and unsocial hours.
Those patterns held independently of algorithmic management. Monitoring is not merely raw material for a later automated decision. It can change working conditions by itself. Watching bodies and movement also creates a different power relationship from logging access to a work system.
A company asking only whether software uses AI may therefore miss the immediate question: what does it observe? Location, keystrokes, camera footage, voice, biometric data and application events carry different risks. Retention periods, access rules and the ability to challenge an inference matter just as much.
Existing law covers only part of the map
The EU's Platform Work Directive provides specific safeguards for people doing platform work. It restricts automated systems from processing private conversations, off-duty data and data used to predict the exercise of fundamental rights. It also provides for transparency, human monitoring and review of significant automated decisions. Member states must transpose the directive by 2 December 2026.
Those protections are substantial, but their platform-work scope does not answer every question raised by evidence from the regular economy. The Quality Jobs Act consultation is considering workplace AI more broadly, including transparent and human-centred automated decisions and protection from excessive monitoring. The eventual proposal is not yet law, and its details remain unsettled.
Legal compliance is only one design input. Employers still have to decide what the system observes, which choices it may influence and where a person can intervene.
Audit the management loop
Workers, managers and procurement teams need to see the whole loop. Five questions reveal more than a generic AI disclosure:
- What is collected? List every input, including location, video, audio, device activity, output metrics and inferred traits.
- What decision does it influence? Separate measurement from recommendations and decisions about pace, shifts, pay, evaluation or discipline.
- Who can contest it? Name a human with authority to review the record, correct bad data and reverse the outcome.
- What happens when signals conflict? Test illness, equipment failure, accessibility needs, shared devices and valuable work that is difficult to count.
- Does control accumulate? Review combinations of monitoring, task allocation and performance scoring, not only each tool by itself.
For a system assembled through api.ish.chat, that means minimising workplace data, keeping decision logs inspectable and placing consequential actions behind explicit human review. In ish.chat, the principle works at a smaller scale: assistance should stay visible as assistance instead of quietly becoming evaluation.
Our earlier article on the labour behind AI provenance argued that technical records should include people, not only models and datasets. Algorithmic management needs the reverse view too. When software acts on people, the record should show who was observed, what rule was applied and who could change the result.
Ask the management question first
The workplace AI debate is bigger than whether a model can write a memo. Software may set the conditions under which that memo, delivery, inspection or shift is completed.
The AIM-WORK studies offer a practical vocabulary. Separate monitoring from management. Separate physical tracking from digital activity logs. Examine direct control of pace and tasks. Look for accumulated exposure. Then ask whether a person can understand and challenge the outcome.
An algorithm can be useful without getting the last word. If a system can alter a shift, evaluation, payment or disciplinary decision, a named person should be able to inspect the evidence and reverse a bad result. That is a better boundary than waiting for the dashboard to become ordinary enough to disappear.
