What are we optimizing when we optimize work?

Photo by Jose Fabula on Unsplash.

Last Thursday, I had the opportunity to moderate a panel discussion on employability, labour rights and technological transformation as part of the First International Meeting on Digital Rights held in Barcelona. The panel featured Luz Rodríguez, Eva Rimbau and José Varela, three complementary voices on a topic that becomes more complex the more closely one examines it.

It was another opportunity to confirm that the great debate about the impact of artificial intelligence on work is not, in fact, primarily a technological debate. It is a debate about how we govern tensions that already existed before AI arrived, but which are now being amplified, accelerated and made harder to balance.

Tensions between the legitimate objectives of companies, the needs of workers and the priorities of society. For decades, in many countries, we have tried to manage these tensions through tripartite frameworks of labour regulation. But the emergence of AI forces us to revisit them.

On the one hand, companies have spent years trying to respond simultaneously to growing demands for competitiveness, productivity, innovation, adaptability and profitability in an increasingly uncertain environment. The pressure to speed up processes, reduce costs and adapt more quickly is not new. But AI promises to take this logic of optimisation to an unprecedented scale.

Workers, meanwhile, are entering this transition after years marked by a feeling of permanent acceleration and a fear of losing employability. In theory, AI should free up time, facilitate lifelong learning and enable richer, more creative jobs. Jobs that, by increasing productivity, could also be better paid. But it can also increase pressure and fear, erode capabilities or create new forms of cognitive dependence.

At the same time, our societies are experiencing growing tensions around inequality, social cohesion, trust in institutions and the difficulty of sustaining some social protection systems. AI is also appearing at a time when many democracies are showing deeply worrying signs of fatigue.

From this broader perspective, it becomes clear how poor it is to frame the scenario in binary terms. The tensions between regulation and innovation, rights and competitiveness, automation and augmentation, efficiency and resilience are not dilemmas, but problems. Most of the goals we pursue are legitimate, but they are very difficult to maximise at the same time. That is the great challenge.

We want competitive, innovative and resilient companies. But we also want sustainable jobs and cohesive societies. We want productivity, but not at the expense of people’s wellbeing. We want automation, but not at the cost of hollowing out learning pathways. We want efficiency, but not organisations so optimised that they end up becoming humanly unsustainable.

AI can help companies, people and governments optimise almost anything. The question is what exactly we are optimising through our decisions, at what cost, and who decides which costs are acceptable.

An organisation that is extremely efficient from an algorithmic point of view may end up being an impoverishing environment from a human point of view. Regulation designed exclusively to protect may end up reducing opportunities for learning or mobility. And a society obsessed with maximising productivity may discover too late that it has damaged the human capabilities on which the sustainability of that productivity depends.

Perhaps that is why it is useful to organise this reflection around three questions: 1) which human capabilities we develop or cease to develop; 2) how power and value are redistributed; and 3) what everyday experience of work we are building.

The AI revolution and the future of human capabilities

Many organisations are using AI to reduce time, speed up processes and eliminate friction. But they sometimes forget that an important part of those “frictions” were also spaces for learning, human interaction and the development of professional judgement. For decades, much professional development took place by carrying out repetitive tasks, observing more experienced professionals or gradually facing increasingly complex problems.

Now, AI is capable of automating many of those entry-level tasks. For example, the junior analyst who no longer builds their first models because a tool generates them automatically; the communications professional who no longer writes initial drafts because an assistant does it for them; the programmer who uses blocks of code without fully understanding the logic behind them; or the HR professional who uses pre-designed questions without having developed their own judgement about how to approach a complex selection interview.

In all these cases, there may be an immediate efficiency gain. But we may also be losing experiences that helped to build professional judgement.

This question is especially relevant in knowledge-intensive professions. In fields such as programming, analysis, design or content generation, AI can enable people with much lower levels of expertise than were required just a few years ago to obtain acceptable results. But that does not necessarily mean we are creating more knowledge or greater collective capability. In some cases, we are simply reducing the need to develop it.

For years, we assumed that the use of advanced technologies automatically raised the skill level of work. But it is becoming increasingly evident that certain forms of AI integration can erode essential cognitive capabilities such as critical thinking, writing, analysis, synthesis or autonomous problem-solving.

The problem is not only that machines are learning more. The problem is that humans may end up thinking less, questioning less, or depending increasingly on systems whose functioning we only partially understand.

Moreover, we may end up thinking in more similar ways. If millions of people use the same tools to summarise, write, analyse, prioritise or decide, there is a possibility that our ways of reasoning will become more homogeneous. AI can broaden access to knowledge, but it can also narrow the diversity of approaches if we all delegate an ever-larger part of our judgement to systems trained on dominant patterns.

And that has profound implications. Organisations need different perspectives to innovate, detect risks and avoid collective errors. Democratic societies need citizens capable of contrasting arguments, sustaining reasonable disagreement and resisting simplifying narratives. If AI ends up weakening those capabilities, the problem will not only be a labour issue. It will be institutional and social.

The AI revolution and the redistribution of power and value

AI does not only transform tasks. It also redistributes value, decision-making capacity and control. And that redistribution does not take place on a blank slate, but on top of inequalities, asymmetries and incentives that already existed before.

Take, for example, the issue of the so-called “data dividend”. Thousands of workers are training, through their daily activity, systems that could potentially replace part of their own tasks. But what happens when people’s accumulated experience feeds systems that reduce the market value of their own capabilities or lower the level of expertise required to access the jobs those people had occupied until then?

Until now, we have tended to interpret data as a business asset. But perhaps we will increasingly begin to see it as the result of a collective contribution distributed among workers, users and organisations. And this could alter some of the traditional balances on which labour relations and corporate pay practices have been built.

Added to this tension is the fact that AI is also beginning to function as a legitimising narrative for certain business decisions.

Announcements of workforce adjustments and collective redundancies in which the automating power of AI is pointed to, more or less explicitly, as one of the causes are becoming increasingly common. In many cases, that explanation will be true. Technology will make it possible to perform with fewer people tasks that previously required larger teams. But in other cases, AI may become a pretext for disguising economic problems, strategic mistakes or cost-cutting decisions that would have happened anyway. In this way, the problem is no longer poor management; AI is to blame.

This nuance matters. Because if we automatically attribute every workforce adjustment to AI, we run the risk of turning technology into an autonomous historical subject, as if companies did not make decisions, when in reality they do. They decide where to invest, which capabilities to preserve, which costs to cut, which deadlines to demand and how to distribute efficiency gains. AI can expand the repertoire of options, but it does not remove the responsibility of choosing between them.

It can also amplify pre-existing inequalities. People with higher levels of education, better networks, greater professional autonomy and greater access to advanced tools may be better placed to turn AI into a lever for productivity, learning and employability. By contrast, those in more vulnerable positions may experience it mainly as a source of control, substitution, intensification or precariousness.

Something similar may happen between companies. Organisations with more and better data, more capital, stronger technological capabilities and greater market power will be better positioned to capture the gains from AI. The rest may become trapped in defensive strategies, forced to adopt technology not so much to transform their value proposition as to avoid falling behind.

That is why the debate about AI cannot be limited to its aggregate impact on productivity or employment. We must also ask who wins, who loses, who decides and who has a real capacity to adapt.

The AI revolution and the everyday experience of work

There is a risk that the productivity gains derived from AI will not translate into equivalent improvements in wellbeing. For decades, part of the social legitimacy of technological progress rested on the idea that greater productivity would eventually generate better wages, less physical effort or more free time. But the recent experience of digital work suggests that this relationship can no longer be taken for granted.

In many environments, technology has not reduced the intensity of work. It has simply made it possible to fill it with more meetings, more messages, more availability, more multitasking, faster response times, and more work compressed into less time. And AI could reinforce this logic of hyper-optimisation, where every efficiency gain is immediately absorbed by new performance demands.

If that happens, we could end up building organisations that are extraordinarily productive, but psychologically unsustainable.

To this we must add the power of control that AI can give employers. We are not only talking about automating tasks, but about monitoring behavioural patterns, inferring performance levels, detecting deviations, predicting risks or continuously comparing people with one another. In certain contexts, this capacity can help prevent problems or improve decisions. But it can also create environments where people feel permanently observed, evaluated and classified.

The result may not be greater engagement, but more anxiety. Not necessarily more responsibility, but more “productivity theatre”. When people perceive that every gesture can be measured, interpreted or used against them, they tend to protect themselves. They avoid risks, hide mistakes, reduce spontaneity and devote more energy to appearing productive than to doing valuable work.

Mental health emerges here as a relevant issue. Not only because of the exhaustion caused by hyperconnectivity or acceleration, but also because of the permanent adaptive anxiety that can arise in a context where the required capabilities are constantly changing and where every technological advance seems to bring us a little closer to the precipice of professional obsolescence.

For years, we have presented employability as an individual responsibility to learn, unlearn, update oneself and reinvent oneself. But when technological change becomes permanent, that demand can become a source of anxiety. One thing is lifelong learning. Quite another is living with the feeling that one never knows enough, that any skill may suddenly expire and that every new tool forces us to redefine our own professional value once again.

In that scenario, lifelong learning ceases to be an emancipatory promise and becomes an endless obligation.

That is why the major question of the coming years will not only be how much employment AI destroys or creates. The question should also be what kind of subjective experience of work it generates.

Because work does not only produce goods and services. It also structures identities, social relationships, learning, self-esteem, a sense of progress and the perception of social usefulness. And all of that is much harder to measure than productivity, but it also has value.

For decades, we have interpreted technological progress fundamentally through the logic of efficiency. But in the current context, that perspective is not only insufficient; it is dangerous. We need to stop focusing on how much work we are capable of automating and focus instead on which human capabilities we want to preserve, which balances of power we want to build and what experience of work we want to make possible.

Because the great challenge of the coming years will not be so much a technological challenge as a governance challenge: how we collectively manage tensions that are becoming more complex and unstable. Not in order to eliminate them — that will probably be impossible — but to prevent them from becoming so unbalanced that they erode what makes our organisations and societies functional and sustainable.

Companies will have to ask themselves not only how to use AI to become more efficient, but also which human capabilities they need to preserve in order to remain innovative and resilient over the long term. They will also need to ask whether their measurement and control systems generate trust or anxiety; whether their automation decisions open opportunities or close development pathways; whether their productivity gains translate into wellbeing or simply into more pressure.

Educational systems, for their part, will have to rethink what it means to prepare someone for an environment in which many technical capabilities can be rapidly automated, but where judgement, critical thinking, contextual creativity and the capacity for human cooperation may become even more valuable.

Meanwhile, governments will have to confront complex debates about the distribution of benefits, social protection, algorithmic regulation, new forms of inequality, rights over data and limits to the power of surveillance at work.

Finally, as individuals, we will have to learn to live with tools capable of enormously amplifying our capabilities without allowing them to completely replace our ability to think, decide and give meaning to what we do.

AI will not decide for us what future of work we will have. But it does force us to decide much more explicitly and consciously what we are willing to prioritise — as companies, as individuals and as a society — and what costs we are willing to assume in order to achieve it.

That is the conversation we need to have, with the seriousness it deserves.

Because the problem, in the end, is not AI. It is what we optimise through our decisions.

+++