A seventeen-year-old student is preparing her first CV. She imagines that a person will read her experience and decide whether she deserves an interview. That may not happen: a system might rank applications, infer skills from keywords or reject an incompatible schedule before anyone intervenes. Another application may later assign her shifts, measure how long tasks take, recommend training and flag who appears likely to leave. The future of work does not begin ten years from now. It is being built today whenever an organisation decides what to turn into data, what to delegate to a machine and which rights an affected person keeps.
Talking about that future with teenagers does not require frightening them or treating every application as an enemy. Systems can remove repetitive tasks, identify hazards or help distribute resources. The key questions are who defines the objective and who bears the cost of an error. Software that arranges schedules to reduce waiting may help; software that ignores education, caring duties or disability converts a human need into a “deviation”. Preparing for algorithmic work means learning to ask concrete questions: which decision is automated, which information is used, who is accountable, and how can it be corrected? Those questions offer more lasting protection than memorising the name of technology that will soon change.
Small decisions manufacture tomorrow’s labour market#
A tool rarely arrives announcing that it will change the relationship between an employer and its workforce. It is introduced as a way to save twenty minutes when selecting candidates, filling a shift or prioritising orders. Yet the accumulation of those choices changes which qualities matter. If speed alone is rewarded, listening to a difficult customer looks unproductive. If permanent availability raises a score, a person with caring responsibilities is penalised. If a filter searches for histories resembling previous hires, it can preserve inequality even though nobody wrote a discriminatory rule. The future is decided through modest criteria which, repeated thousands of times, ultimately define a profession.
A tool rarely arrives announcing that it will change the relationship between an employer and its workforce.
That is why prediction must be separated from decision. A model may estimate that an application resembles earlier profiles, but it does not know who will learn best in a new team. It may forecast demand, but cannot determine by itself which shift change is compatible with a family. Prediction is an input; a decision contains values, consequences and responsibility. When a company presents an output as inevitable, it erases these dimensions. Good practice requires naming the accountable person, explaining which factors mattered and allowing time to consider context. “The system recommends it” should never close a conversation affecting income, health or reputation.
What a young person needs before a first job#
A family can prepare a young person without inspecting every algorithm. The first skill is documentary: keep the job offer, conditions, shift-change messages and significant evaluations. The purpose is not to live defensively but to have facts when a contradiction appears. The second skill is linguistic: replace “the algorithm dislikes me” with verifiable questions. Which data were used? Which criterion produced this score? Who has authority to review it? What happens during review? A precise request makes a useful response more likely, helps uncover inaccurate records and prevents a legitimate concern from being dismissed as an impression that nobody can test.
The third skill is setting limits on data. A workplace application may request location, camera, contacts, physical activity or permanent access to a device. The presence of a permission screen does not prove necessity. Before accepting it, a young worker can ask which specific task needs it, during which hours, how long it is kept and whether an employer-owned device is available. The Spanish Data Protection Agency’s guide to data protection in employment relationships explains principles of information, necessity and proportionality. A family can help with reading, but the conversation with the employer should strengthen the young person’s autonomy rather than replace their voice.
Young workers must also learn not to “work for the score”. When a dashboard displays times, stars or rankings, it is easy to confuse the indicator with the craft. An apprentice needs to know that asking for help, stopping unsafe work, recording an incident and serving a person carefully remain part of the job even when they earn no points. They should ask a supervisor how quality and exceptions are evaluated, not only quantity. If the only answer is a number without context, there is an organisational risk. No young person should believe that obeying an obviously unsafe instruction is professional simply because a screen marked it urgent.
The map of power behind a neutral-looking screen#
Evaluating a system requires a map of power. Who buys the tool? Who configures thresholds? Who supplies the data? Who receives the alert? Who suffers the consequence? Who can stop it? These are often different actors. Human resources may contract the service, a supplier may update the model, a manager may use the score and a worker may lose hours. When nobody connects the journey, each actor can claim to be performing only one small part. The map restores accountability: it reveals where information, oversight, training and a genuine power to correct are needed.
Human oversight is not a signature placed after automatically accepting hundreds of recommendations. A person supervises when they understand the purpose, receive additional information, have time to doubt and can alter the outcome without punishment. They also need to record why they changed it so repeated failures become visible. If the organisation measures the supervisor by how rapidly recommendations are confirmed, the human becomes a button. An audit should examine modification rates, reasons, affected groups and consequences rather than merely confirming that somebody formally appears in the process.
Transparency does not mean publishing thousands of lines of code. An affected person needs a functional explanation: what decision was made, which categories of data mattered, which source might contain an error, what effects follow and how review can be requested. In Spain, Article 64.4(d) of the Workers’ Statute recognises legal worker representatives’ right to information about the parameters, rules and instructions behind algorithms or artificial-intelligence systems influencing decisions capable of affecting working conditions. Information should enable participation, not become an incomprehensible formality.
Three futures for the same tool#
Imagine a system forecasting how many workers a shop needs. In the first future, the employer pursues cost adjustment alone: shifts are published at short notice, hours change automatically and refusals are penalised. The prediction may be accurate, but all uncertainty is shifted to workers. In the second future, the system proposes schedules respecting declared availability, rest, education and caring duties; a person resolves conflicts and changes are justified. In the third, the team participates in setting rules and evaluates whether the tool actually reduces workload. The technology is similar; what changes is the distribution of risk and the possibility of intervention.
Now consider a recruitment tool. Poor use turns an opaque score into an automatic gate. Less harmful use limits it to organising administrative information and prohibits automatic rejection. Better design reviews necessary competencies, offers disability adjustments, tests outcomes across groups and documents every exclusion. The European Union Artificial Intelligence Act classifies certain systems used for employment, recruitment and worker management as high-risk and creates duties depending on the actor and applicable timetable. The lesson is not to wait for regulation to solve everything; it is to design verifiable, reversible processes now.
The third example is performance assessment. If a system counts keystrokes, calls or deliveries, people will optimise whatever is counted. Cooperation may decline, problems may be concealed and learning may worsen. An alternative combines limited indicators with professional observation, dialogue and the right to provide context. No metric should trigger a serious penalty by itself. Before implementation, the employer should identify behaviour the indicator could induce: skipping breaks, avoiding difficult cases, competing with colleagues or creating artificial activity. Sound design asks not only whether measurement predicts something, but also which behaviour it creates.
A practical test before automating a decision#
The first phase is to write down the problem without mentioning artificial intelligence. “We take too long to communicate schedules” permits comparison between options; “we need AI for scheduling” has chosen a solution before understanding the need. Establish a baseline next: notice periods, involuntary changes, errors, correction time, satisfaction and differences between groups. Success must include human constraints, such as not reducing notice or increasing mental workload. If financial savings are the only measure, the system will learn to transfer costs towards people with the least power to refuse them.
The second phase tests difficult cases before real people are affected. Include missing data, similar identities, poor connectivity, disability, an unusual career history, a family emergency and changes in demand. For each case, ask what the system proposes, what the responsible person sees and how long correction takes. The trial should produce a failure log and an explicit choice: repair, restrict the use or abandon it. Buying a licence does not create a duty to deploy it. The ability to say “this does not work in our context” is an organisational competence, not a failure.
The third phase establishes an effective appeal. A person knows where to complain, receives confirmation, understands the deadline and can reach somebody with authority. While review occurs, irreversible consequences should be avoided whenever possible. Protection against retaliation matters too: nobody should lose opportunities for questioning a record. The European directive on platform work contains specific rules on algorithmic management, information and human review within its scope. Even outside platform work, those elements provide a useful practical reference: an identifiable decision, a reachable owner and a review capable of genuinely changing the outcome.
The fourth phase sets an expiry and review date. Data, teams and economies change; a useful tool can deteriorate. Every system needs an owner, periodic review, harm indicators and a withdrawal rule. Complaints, corrected errors, differences between groups, turnover, accidents and newly created hidden labour should be examined. New uses must also be challenged: an application bought for planning should not quietly become a disciplinary mechanism. A changed purpose requires a fresh evaluation. An absence of recorded incidents proves little when the reporting channel is unknown or makes people afraid.
Family conversations that prepare without creating fear#
A good conversation can begin with a situation rather than a lecture: “If an application cuts your hours and nobody can explain why, what would you do?” The young person proposes steps; the family helps order them: save the message, ask for the criterion and data, request review and seek support if the consequence is serious. A short, respectful email can be rehearsed. It is also useful to distinguish a poor decision from illegality: not every automated error breaches a rule, but every significant outcome deserves an accountable person and a correction mechanism. Important employment or data-protection matters may require appropriate professional or union advice.
Families can recognise abilities that no dashboard summarises well: caring, explaining, improvising safely, learning from a mistake, listening and collaborating. Helping a young person name these qualities gives them a richer vocabulary than “being productive”. When preparing for an interview, they can describe concrete situations and ask how the role is evaluated, which tools are used and who reviews results. Asking does not guarantee a perfect employer, but it reveals culture. A clear answer about goals, limits and support demonstrates more maturity than a promise of total objectivity based on technology.
Adults should apply the same reasoning to their own workplaces. If we accept intrusive monitoring because “that is how things are”, we teach resignation. If we reject every use of data, we lose opportunities to improve. The useful example is deliberation: identify benefit, risk, alternative, owner and review. Sharing a real experience with children —a poorly assigned shift, an incomplete metric or helpful automation— turns an abstract concern into civic literacy. The aim is not for everybody to program models, but for nobody to surrender the right to ask who decided and under which rules.
Choosing a dignified future of work today#
Employers, governments, unions, schools and families are writing the social contract of automated work. An employer contributes when it permits appeal; a school when it teaches documentation and argument; a government when it inspects; and a family when it protects without removing autonomy. Effective rules combine innovation with boundaries: minimum data, a specific purpose, impact assessment, participation by people doing the work, oversight with power and remedy when something fails. Dignity is not a brake attached at the end. It is a quality condition: systems that ignore context produce poorer decisions and organisations less able to learn.
Model names may change within a few years, but the questions will remain. What are we optimising? Which part of the craft cannot fit inside a number? Who is excluded? Who can stop the decision? What happens when it is wrong? A worthwhile future is neither one where a machine decides everything rapidly nor one where every technological aid is rejected. It uses tools to expand capability without reducing people to profiles. That future is being chosen today in every contract, permission, metric, explanation and appeal that we accept as normal.
