Two Kinds of Talent, and the Quality Both Need
A specialist or a generalist can take very different paths. Neither can afford to wait for someone else to decide what happens next.
Larry Pang
Co-founder, Lantr
August 12, 2026
6 min read
In a recent conversation about AI, education, and work, MasterClass founder David Rogier offered a sharp prediction: the workforce may develop a barbell shape.
At one end are genuine specialists, among the very best in a real field and difficult to replace. At the other are high-agency generalists, people who can step into different kinds of problems, make judgments, and move without waiting for detailed instructions. Rogier believes the broad middle will become less secure.
Stanford professor and ImageNet co-creator Fei-Fei Li agreed with the broad shape of the argument. Their discussion kept returning to one quality both groups need: agency.
Agency can sound abstract. Here it means something quite practical: noticing that a problem is worth solving, deciding what to do, and continuing to move the work forward when nobody has prepared the next step for you.
Whether specialist or generalist, neither can keep waiting to be told what to do next.
Now consider an ordinary student's week.
Someone sets the timetable.
Someone assigns the homework.
Someone organizes the test preparation.
Even the summer project often arrives with its topic, steps, and deadline already chosen.
How often does that week ask the student to notice a problem nobody assigned, choose a direction, and take responsibility for finishing? Usually, not very often.
This is not a criticism of students or parents. They are responding sensibly to the system around them. The problem is that a tightly organized path can produce strong performance while leaving very little room to practice making unassigned decisions.
Agency is hard to teach in a standalone class and impossible to prove with a line on a résumé. It becomes visible in what a student chooses, what they do when the first plan fails, and whether anything real exists at the end.
AI does not make knowledge optional
When AI can answer almost any question on demand, it is natural for parents to wonder whether learning still matters in the same way. But more answers do not remove the need for knowledge. They increase the number of answers a student must judge.
Knowledge now does two jobs. It helps a student form an answer, and it helps them recognize when a ready-made answer is unreliable, irrelevant, or simply beside the point.
A polished explanation can still cite the wrong study. Working code can still contain a security problem. A confident market analysis can still begin with a false assumption. Without enough subject knowledge, a student may not know where to be skeptical.
What can AI genuinely help a student do?
Ask without embarrassment
A student can request another explanation, a simpler example, or immediate feedback without worrying that the teacher will lose patience.
Test an idea sooner
With a small enough scope, a beginner can make a first prototype before mastering every technical prerequisite.
Spend time on better questions
When AI handles part of the formatting, debugging, or initial research, the student has more time to decide what to build and whether it actually works.
The benefit is not that the student can avoid learning. It is that they can reach the consequential decisions sooner.
How can you tell whether the student really understands?
This leads to a question parents ask often: Does using AI count as cheating?
The word “cheating” is too blunt for most work outside an exam. A more useful question is whether AI extended the student's thinking or replaced it.
Ask the student to explain the project to someone who does not know it and will not be easily impressed. Can they explain what problem it solves, why they made each important decision, where the first attempt failed, and what they would change next?
A student who owns the work can usually answer in specifics. They remember the awkward first version, the suggestion from AI that led nowhere, and the moment a user did something they did not expect. A student who does not own it tends to retreat into a tour of the features.
Do not ask only what the student made. Ask what changed in their mind while making it.
A useful project usually includes a few simple steps:
Find a small problem that a real person actually has.
Build the smallest version that person can try.
Watch where they become confused or stop using it.
Change the product in response, then put it online.
Explain every important decision in your own words.
A mentor should not take over the work
In our work with students, projects tend to stall in three familiar places:
The idea is far too large
The student sets out to build “an AI healthcare platform,” circles the concept for weeks, and never reaches something another person can try.
The first obstacle stops everything
One technical problem they cannot solve alone becomes the point where the project disappears into a folder.
It is nearly finished forever
The project runs on the student's laptop, but there is no link, no user, and no evidence that it works outside the room.
A good mentor helps reduce the scope, removes a blockage without seizing the decisions, and insists that “almost ready” eventually meets a real user. The mentor provides structure; the student still has to own the judgment.
Parents can use the same boundary. Help the child find a smaller first step. Ask questions when they are stuck. Resist the urge to rewrite the plan, fix every rough edge, or turn the project into the version an adult would have made.
Launch is where the next part of learning begins
Once a project reaches real users, the student encounters feedback no rubric could have predicted. Someone misunderstands the first screen. A feature they were proud of goes unused. A problem they considered minor turns out to matter most.
That experience is valuable because reality does not provide a model answer. The student must decide which feedback matters, what to change, and when to keep a decision despite disagreement.
A young person may eventually become a specialist or a generalist. Either path can begin with one complete act of creation: notice a problem, choose a direction, work through setbacks, and let the result meet the world outside their own head.
Agency is not a label. It is the habit of moving work forward when no answer key tells you what comes next.
Source note: This essay responds to a published conversation hosted by Marina Mogilko with Fei-Fei Li and David Rogier. Rogier introduced the barbell hypothesis; Li agreed and emphasized agency. The explanations and applications to student projects are our own synthesis. Neither guest is affiliated with Lantr.