What’s Worth Learning in the Age of AI?
I gave a talk at TEDxEsade 2026 called What’s Worth Learning in the Age of AI? It is a question I keep coming back to as AI takes on work that once required years of training.
If a machine can write the code, prepare the report, and explain the result, where should we put our effort? What should a student study? What should an experienced professional learn next?
My answer starts with a change in what we consider valuable. Producing something is becoming easier. Deciding what deserves to exist, understanding how it works, and making it useful still demand a great deal from us.
We need to learn how to take responsibility for what we can now create.
Learn to build something worth using
Consider the sandcastle below. Four words frame the work of building it. Context, Questions, Plan, Build. They describe a way of working with AI that begins before the first prompt asking it to produce anything.
Context means explaining the situation: who has the problem, what they do today, and what constraints matter. Questions expose the gaps in that understanding. A plan gives us something to challenge before we invest in building.
Imagine a team copying information between emails, spreadsheets, and a business application. AI may help build a tool that connects that work. But the first question is what makes the current process painful. Where do errors happen? Who needs to approve a change? What would make someone trust the new tool?
Without those answers, we can build an impressive solution to a problem nobody has.
And a working demonstration is only the beginning. People have to adopt it. It has to fit their workflow. It has to create value, whether that means time saved, lower costs, more revenue, or better quality.
Value = f(Time, Tokens, Judgment, Adoption)
Think of it as a reminder, rather than a calculation. AI usage consumes resources. Human judgment shapes the result. Adoption determines whether anyone benefits. A beautiful tool that nobody uses has failed an essential test.
Learn to direct your attention
When AI can work on several tasks, the temptation is to start everything at once. We can ask one agent to research, another to build, and another to review. Soon we are the busiest person in a room full of tireless assistants.
Managing several AI tasks means keeping their contexts distinct and deliberately choosing where our attention belongs.
Each task needs a clear goal and constraints. Sometimes we need to think about the overall design. Sometimes we need to inspect a detail or make a decision that unblocks the work. Eventually, the separate pieces must work together.
Starting more tasks is easy. Knowing which one needs you now takes practice.
Learn enough to understand what you approve
Imagine receiving this update from an AI assistant.
“All tests have passed. The database migration and recovery checks are complete. Shall I deploy the changes?”
It sounds reassuring. “What am I actually approving?” is the question I would ask before agreeing.
This is an uncomfortable question because fluent technical language can make us feel informed before we understand anything.
Suppose an AI assistant helps you build a company intranet. It asks about deployment, access, or indexing. If those words mean little to you, a casual “yes” can authorize something very different from what you intended. A system containing internal information might become publicly accessible.
You do not have to become an expert in every subject before you begin. You do have to understand the decisions you are making.
Tell the AI your starting point. Ask it to explain unfamiliar terms in plain language. Ask what a proposed action changes, who it affects, and how you can check the outcome. When you cannot judge the answer, bring in someone who can.
Building can become a way of learning, provided we stay curious at the moments when it would be easier to click “approve.”
Learn to notice when there is nothing on the plate
AI can arrange the table beautifully. Our job is to make sure there is actually something on the plate.
A polished report may contain no decision. A long email may leave the reader unsure what to do. A confident article may repeat familiar ideas without adding a point of view.
Before sending something, ask: What does this help the reader understand or do? Would I want to receive it? Have I made the reader’s work easier?
And when a message is personal, put yourself into it. Your context and care matter. Delegating the wording should not mean delegating the relationship.
Learn when to doubt a convincing answer
An AI system can produce an enthusiastic assessment of a business idea. That does not establish that a customer wants it.
Agreement feels good. It can also keep us from asking the question that matters. What would make this fail?
Ask for the opposing case. Look for missing context and assumptions you have quietly accepted. Consider whether the task is too broad or ambiguous to trust a single response. Then test the idea against something outside the conversation: a customer, an experiment, an observable result.
The ability to get an answer is useful. The ability to decide what would count as evidence is essential.
Learn when to delegate and when to think together
The centaur and the cyborg describe two ways of working with AI. These metaphors also appear in research on how people work with generative AI.
The centaur divides the work. You delegate a well-defined task, such as comparing documents or preparing a first report, and keep responsibility for the goal and the review.
The cyborg works through short, continuous exchanges. You draft, question, revise, and build together. The conversation helps you discover what you think as well as produce the result.
Both approaches are useful. The skill is recognizing which one the work needs, and when to switch.
As AI expands what we can do, what will we choose to become better at?
My answer is to practice building useful things, asking harder questions, and understanding the consequences of our decisions. The more AI enables us to create, the more those abilities matter.
If you could build almost anything, how would you decide what was worth building?
