Gavin Fudge

11/05/2026

8 min read

If AI is doing all the work, what are humans doing?

Something is happening to knowledge work.

Calculate. Design. Write. Analyse. Execute.

These activities once defined expertise. Organisations depended on people who could perform complex tasks with skill and reliability.

Now powerful AI systems can complete many of those tasks in seconds. Work that only a few years ago required hours of focused effort can now be produced almost instantly.

The increasing presence of AI in everyday work is quietly shifting the fundamental value - and meaning - of what we do. The change is subtle, but it is beginning to ripple through organisations everywhere.

Which raises an interesting question:

If AI is doing the work, what exactly are humans meant to do?

The emerging answer is both simple and profound: humans provide judgement.

Skill and judgement are not the same thing

For a long time, professional expertise bundled two things together: skill and judgement.

Skill is the execution layer of work. It includes activities such as analysing data, drafting reports, modelling scenarios, generating designs, and synthesising information. These tasks often involve complex reasoning, but they follow patterns. With enough examples and enough computing power, machines can learn to reproduce them.

This is precisely what modern AI systems excel at. Large language models and other machine learning systems process enormous volumes of data, identify patterns, and generate structured outputs at remarkable speed. Tasks that once required trained specialists can now often be performed by machines with impressive competence.

Judgement, however, is something different.

Sir Andrew Likierman of London Business School describes judgement as the ability to combine relevant knowledge and experience with personal qualities to reach a decision when rules alone are insufficient. It requires context, awareness of uncertainty, and ultimately accountability for the outcome.

Judgement is what happens when a situation refuses to fit neatly inside a model. It is the act of deciding what matters, what can be trusted, and what should be done when the available information is incomplete.

“Prediction, not narration, is the real test of our understanding of the world.”

‍Nassim Nicholas Taleb, The Black Swan: The Impact of the Highly Improbable

The AI paradox

The rapid rise of AI has produced a curious paradox.

The more capable machines become at performing skilled work, the more valuable human judgement becomes. AI expands the space of possible answers, but humans still decide which answers make sense.

A model might generate ten strategic options, but someone still must decide which one aligns with reality. An AI system might analyse thousands of radiology images with extraordinary accuracy, yet a doctor must interpret what that result means for a particular patient. A generative system might produce a polished strategy document in minutes, but a leadership team must still decide whether the strategy is wise.

The machine produces possibilities. Humans apply judgement.

The black swan problem

One of the reasons judgements remain so important is that the world is not purely statistical.

In his book The Black Swan, Nassim Nicholas Taleb describes how rare, unexpected events shape history far more than predictable trends. These events are difficult to foresee precisely because they fall outside the patterns that past data suggests.

Statistical systems - including modern AI models - are very good at identifying patterns in historical data. But by definition, they struggle with events that do not resemble anything that came before.

In Taleb’s metaphor, most systems assume that all swans are white because every swan they have seen before has been white. The appearance of a black swan reveals the limits of that assumption.

Large language models operate in a similar way. They are trained on enormous datasets of past information, which makes them extraordinarily effective at recognising patterns and generating plausible responses. But their understanding of the world is still fundamentally probabilistic. They are designed to produce what is most likely - not necessarily what is true in a new and unfamiliar situation.

Judgement is what allows humans to recognise when the model’s assumptions may no longer hold. It is what allows someone to ask the uncomfortable question: what if the situation is different this time?

Without judgement, automation can easily turn past patterns into future mistakes, and today’s solution becomes tomorrow’s problem.

Where we are in the AI timeline

Understanding why judgement matters also requires a brief look at where AI currently sits in its development.

Most of today’s systems fall into the category researchers call Artificial Narrow Intelligence (ANI). These systems are highly capable within specific domains - language generation, image analysis, code creation, and specialised reasoning tasks. They are powerful, but they are still bounded.

Beyond ANI lies the concept of Artificial General Intelligence (AGI): systems capable of reasoning flexibly across many domains in a way comparable to human cognition. Beyond that lies the speculative horizon of Artificial Superintelligence (ASI), systems whose cognitive abilities surpass human intelligence across most fields.

Exactly when these milestones might arrive remains uncertain. Surveys of AI researchers show wide variation in expectations, though many place the possibility of broadly human-level AI capabilities sometime around the middle of this century.
(AI Impacts Expert Survey)

Major AI laboratories themselves now openly discuss the need to prepare for increasingly general AI systems in the coming decades.
(OpenAI – Planning for AGI and Beyond)

The exact timeline matters less than the direction of travel. AI capability is improving quickly, and the first layer of work it affects is skill.

The risk of moral deskilling

Philosophers studying technology have raised an important concern about this shift.

Shannon Vallor describes the danger of moral deskilling - the gradual erosion of human judgement when decision-making becomes overly delegated to machines. When people rely too heavily on automated outputs, they risk losing the habit of evaluating situations independently.

This idea echoes an earlier warning from the philosopher Günther Anders, who wrote about the tendency for humans to defer responsibility to machines simply because machines appear authoritative.

We already see hints of this behavior today. AI systems present outputs confidently and often with convincing explanations. Yet anyone who works with them regularly knows that hallucinations, subtle errors, and misleading conclusions are still common. Without human judgement, those mistakes can easily slip through unnoticed.

Judgement is not a decorative addition to AI systems. It is the safety mechanism.

Judgement is becoming a structural skill

Historically, judgement was associated with seniority. Managers decided and everyone else executed.

But when machines begin performing much of the execution layer of knowledge work, that hierarchy begins to shift. Judgement becomes necessary at every level.

Someone using AI to analyse customer behaviour must judge whether the patterns reflect real market behaviour or statistical artefacts. Someone generating code with AI must judge whether the architecture is robust. Someone drafting policy with AI must judge whether the logic holds up under scrutiny.

The future workplace may not be defined primarily by who has the most technical skill. Increasingly, it may be defined by who applies the best judgement.

From tools to agents

There is another shift beginning to unfold beneath the surface.

For the last two years AI has largely appeared as an assistant - something that generates content or answers questions when prompted. But increasingly these systems are evolving into something else entirely.

Agents.

AI agents are systems that pursue goals. They can plan tasks, gather information, interact with other software tools, and complete sequences of work autonomously. This changes the conversation entirely.

Once AI systems begin performing multi-step tasks independently, someone must decide what the agent’s goal should be, what data it can access, what rules it must follow, and when it should escalate to a human.

Those decisions are not computational - they are judgements.

The real skill of the AI era

If machines increasingly perform skilled execution and humans increasingly apply judgement, the most valuable capability may become something slightly different.

Not prompting. Not coding.

But designing systems.

The ability to structure problems, define goals, and build processes where humans and intelligent machines collaborate effectively.

There is already a name for the thinking style behind this.

Computational thinking.

And it may become one of the most important intellectual tools of the AI era.

‍

References

Likierman, A. (2020). The Elements of Good Judgment. Harvard Business Review, 98(1), 102-111.

Taleb, N. N. (2007). The Black Swan: The Impact of the Highly Improbable. Penguin.

AI Impacts. (2023). 2023 Expert Survey on Progress in AI. Available at: https://wiki.aiimpacts.org/ai_timelines/predictions_of_human-level_ai_timelines/ai_timeline_surveys/2023_expert_survey_on_progress_in_ai

OpenAI. (2023). Planning for AGI and beyond. Available at: https://openai.com/index/planning-for-agi-and-beyond/

Vallor, S. (2016). Technology and the Virtues: A Philosophical Guide to a Future Worth Wanting. Oxford University Press.

Anders, G. (2025 ed.).The Obsolescence of the Human. University of Minnesota Press.

‍

The era of judgement

What happens when your pre-order campaigns run from real data

A critical path that updates itself

Explore More