The most common debate about AI and work is framed as a replacement question: which jobs will AI eliminate? It is a legitimate question, but not the most useful one for someone who wants to act intelligently in the present. The more useful question is different: what skills remain valuable when AI does what it can do, and what is worth learning now to be well positioned?
The wrong question
“Will AI take my job?” assumes that jobs are monolithic blocks that can be fully replaced. In reality, jobs are bundles of tasks, and AI is better at some of those tasks than others.
The historical evidence from previous waves of automation — the printing press, the calculator, computers, the internet — shows a consistent pattern: technology eliminates specific tasks within jobs, which transforms jobs more than eliminates them, and creates new categories of work that did not previously exist.
Accounting did not disappear with spreadsheets: it was transformed. Accountants used to spend 80% of their time on manual calculations; now they spend that time on analysis and interpretation. The value output increased; repetitive tasks decreased.
AI is producing a similar transformation, but faster and across more sectors.
What AI is already changing
Standard text generation tasks. First drafts, reformulations, summaries, translations, follow-up emails, status reports. AI can do this faster and with sufficient quality for many cases. Human time spent on these tasks is falling in a documented way.
Information processing. Classifying, labelling, extracting data from documents, summarising. Tasks that previously required hours of manual work can be executed in minutes with AI pipelines.
Repetitive code. Unit tests, documentation, boilerplate, standard functions. AI does not replace the software engineer, but eliminates the most tedious part of their work.
Basic data analysis. SQL queries, standard visualisations, descriptive statistics. The barrier to this type of analysis has fallen significantly.
Information search and synthesis. Initial research, option comparison, mapping an unknown field. What previously took days of reading can be condensed into hours.
What remains human
The capabilities that AI has the most difficulty replicating — and that will likely remain valuable in any near-future scenario:
Deep contextual judgement. Understanding the power dynamics in an organisation, reading the real motivations behind a negotiation, calibrating what can be compromised and what cannot in a specific relationship. This requires situated experience that AI cannot have.
Relationships and trust. A client trusts a specific person, not the language model that person uses. Trust is built over time through consistency, accountability and understanding of personal context. It is not delegable.
Real accountability. AI produces recommendations; someone has to sign the contract, make the decision, take responsibility for the consequences. Accountability remains human because consequences fall on people.
Creativity with judgement. AI can generate variants and combinations of what already exists. Genuinely new ideas — those requiring seeing something nobody has seen before or rejecting established consensus — remain rare in AI outputs.
Leadership and influence. Motivating a team, managing conflict, creating a work culture, convincing people who are not convinced. All of this requires presence, empathy and understanding of what moves people.
Skills for the AI environment
It is not possible to predict with precision what skills will be most valuable in five or ten years, because the pace of change makes specific predictions expire before you can act on them. But there is a set of capabilities that are robust in practically any scenario:
Using AI with judgement. Knowing what to delegate, how to verify, when to trust and when to doubt. This is the most immediately useful skill and the most underestimated. The difference between someone who uses AI effectively and someone who does not use it — or uses it badly — is already visible in many professional environments.
Critical thinking and output evaluation. AI produces a lot of content. Knowing how to distinguish what is correct, what is plausible but incorrect, what is well formulated but irrelevant, requires analytical capability that AI cannot self-evaluate.
Precise communication. Prompts are instructions. Effective prompts require clarity, specificity and the ability to articulate exactly what you want. Precise written communication — always valuable — is even more valuable when it is the interface with the most productive systems available.
Deep domain knowledge. AI knows a lot about everything in general, but deep knowledge of a specific domain — its nuances, its exceptions, its historical context — remains the criterion for evaluating whether AI outputs are correct. The expert who also knows how to use AI is significantly more valuable than the expert who does not use it, or the AI user without domain expertise.
A productive relationship with technology
The most useful attitude towards AI is neither uncritical enthusiasm nor defensive scepticism. It is the same attitude one should have towards any powerful tool in history: curiosity, active learning, judgement in application.
AI is already here. Ignoring it has a growing cost. Using it without judgement has a different cost: mediocre outputs presented as your own, errors that go undetected, dependence that is not calibrated against the real limits of the tool.
What this course has attempted to build — from what AI is to how agents work — is not technical knowledge for its own sake. It is the conceptual foundation for making informed decisions: knowing when to trust, when to verify, when to delegate and when not to.
The productive relationship with AI is the same as with any high-leverage tool: understand it well, use it where it contributes, maintain judgement where it matters. That does not change with each new model version. It is what makes today’s learning still useful tomorrow.
