Many conversations about AI start with what it will replace. A more useful question is what becomes more valuable when AI handles the work it does well.
If AI can generate content, what happens to the value of taste? If it can process information across systems, what happens to the value of connecting it? If it can produce options at scale, what happens to the value of choosing well? If it can optimise towards a goal, what happens to the value of choosing the goal? If it can handle routine exchanges, what happens to the value of trust? The pattern is reallocation. Skills that mattered when work was mostly execution give way to skills that matter when work is mostly direction, curation and relationship.
What the evidence says
The World Economic Forum's Future of Jobs Report 2025, based on responses from more than 1,000 employers representing over 14 million workers, expects 39% of workers' core skills to change by 2030. Analytical thinking remains the most sought-after core skill, named by 69% of employers, followed by resilience, flexibility and agility, and leadership and social influence. AI and big data are the fastest-growing skills. The human and the technical rise together.
Economists have described the mechanism for two decades. Autor, Levy and Murnane's task model separates routine work, which follows explicit rules and can be codified, from non-routine work that needs judgement, interpretation or interpersonal skill. Each wave of technology has taken on more routine tasks and raised the value of the non-routine ones around them. Agentic AI moves the boundary further, into tasks once thought too varied to automate. Value moves with it, toward the capabilities that remain hardest to codify.
Erik Brynjolfsson calls the choice between these paths the Turing Trap. Technology designed to imitate and replace people tends to concentrate value and stall wages. Technology designed to augment people creates new capabilities and shares the gains. The organisations getting the most from AI are designing for augmentation, and McKinsey's 2025 State of AI survey supports this: the highest performers were nearly 3 times as likely as others to have redesigned their workflows around AI.
Five capabilities stand out.
Taste: knowing what fits
AI produces far more than any team could. Volume is not value. Value comes from choosing the output that fits this context, this audience, and this moment. Taste, in this sense, is recognising fit: whether a reply matches the tone the brand wants, whether an analysis answers the question that matters, whether a recommendation reflects how the organisation wants to act.
Taste cannot be delegated, because it rests on values and intentions that live outside the system. It can, however, be built into a system's boundaries. A financial services system that explains products but never advises expresses taste. So is a customer system that keeps its tone during a complaint, or a configured sales system that tells a buyer when a product is not the right fit. People who understand what the organisation is trying to be make these decisions. The system then applies them consistently, thousands of times a day.
As generation becomes universal, taste becomes a differentiator. Everyone has the same models. Not everyone knows what to do with them.
Context synthesis: connecting across functions
AI handles tasks well within a function. It processes this document, analyses that dataset, answers this enquiry. It is less good at seeing how a pattern in operations connects to a pattern in customer behaviour, or how a change in one area ripples through another.
That connecting is context synthesis. It has always been valuable, but it was often crowded out by the time spent compiling information. When a manager spent hours gathering reports from different systems, little time remained to think about what they meant. Once compilation is automated, synthesis becomes the main contribution.
Systems help most when they surface cross-functional patterns for people to interpret. A system can show that reply times relate to conversion, that certain product configurations cluster in certain customer segments, or that exceptions vary by channel and season. Deciding what those patterns mean, and what to do about them, stays with people.
Judgement: choosing among options
When AI can produce many good options, the scarce skill is choosing. Judgement weighs what the data cannot fully capture: risk appetite, relationships, timing and second-order effects.
Judgement also has a quality problem that is easy to overlook. Kahneman, Sibony and Sunstein's research on noise shows that experienced professionals often reach different conclusions on the same case, and the same person can decide differently on different days. That variation is costly and usually invisible.
AI can help here without taking the decision away. A system that presents options with their reasoning, shows how comparable cases were decided before, and flags what is uncertain gives people a more consistent basis for judging. The decision stays with the person accountable for it. It simply becomes better informed and more consistent across the team.
Strategic instinct: questioning the objective
AI optimises towards the objective it is given. Someone still has to ask whether it is the right objective. A system can make quoting faster. Only people can decide whether the business should be quoting that kind of work at all, at that price, for those customers.
Richard Rumelt describes good strategy as starting with a diagnosis: a clear account of what is going on and what matters most. Optimisation can follow a diagnosis. It cannot make one. As optimisation becomes easy, framing the problem becomes the strategic skill, and the people who can see when the goal itself needs to change become more valuable.
Trust: building relationships
AI can draft messages and handle transactional exchanges well. Trust is built between people, over time. Mayer, Davis and Schoorman's model of organisational trust names three foundations: ability, benevolence and integrity. Systems can demonstrate ability by being reliable and accurate. People show benevolence and integrity in how they handle a difficult moment, keep a promise, or say no when no is the honest answer.
When systems handle routine exchanges well, people have more time for the conversations that build trust, and those conversations carry more weight. A relationship manager freed from status queries can spend that time on the client who is worried, or the one who is about to grow.
What this means for how work is designed
Roles change shape. Jobs that combine execution with judgement keep the judgement and gain time for it. Three design choices make the difference:
Route decisions, not just tasks. Design workflows so each decision reaches the person best placed to make it, with the context they need.
Measure decision quality, not only speed. Track accuracy, consistency and outcomes, so better judgement is visible and rewarded.
Train for the new work. Review, exception handling and governance are skills. They need practice and support, like any other.
Signs an organisation is ready for the shift
Senior people spend more time reviewing and deciding than assembling.
Teams can explain why a system's output is right or wrong, not only whether it is.
Patterns across functions are discussed, not only reported.
Roles have been redesigned around judgement, with training to match.
Success is measured by decisions and relationships, not only throughput.
How Mitochondria's agentic AI systems support people's judgement
The approach. Each unit is designed to handle execution and to bring decisions to people with the context they need: the inputs, the reasoning, comparable past cases and what is still uncertain. Boundaries set with the organisation encode its taste and its risk appetite.
What it enables. People spend their time reviewing exceptions, patterns across functions, and relationships. Decision memory keeps the reasoning behind each choice, so judgement becomes more consistent and builds across the team rather than staying with one person.
The impact. The organisation gains capacity, and its people's work becomes more about the capabilities that grow in value. A companion piece covers how to build these skills in teams.
Read more about Mitochondria's approach.
Sources
FM Magazine, analytical thinking remains top core skill for employers
Autor, D. H., Levy, F. and Murnane, R. J. (2003). The skill content of recent technological change. Quarterly Journal of Economics, 118(4).
Brynjolfsson, E. (2022). The Turing Trap: The Promise and Peril of Human-Like Artificial Intelligence. Daedalus, 151(2).
Kahneman, D., Sibony, O. and Sunstein, C. R. (2021). Noise: A Flaw in Human Judgment.
Rumelt, R. (2011). Good Strategy Bad Strategy.
Mayer, R. C., Davis, J. H. and Schoorman, F. D. (1995). An integrative model of organisational trust. Academy of Management Review, 20(3).
—
Mitochondria builds agentic AI for evidence-led core operations. Research-led and model-agnostic. ISO/IEC 27001:2022 certified, with controls aligned to the GDPR, UK GDPR, India's DPDP Act and Australia's Privacy Act. Amsterdam and Pune. Request a working session →