There is an old constraint hidden inside nearly every form of education: good practice is expensive.
A teacher can listen to only so many presentations. A coach can review only so many performances. A mentor can sit through only so many rehearsals. Feedback is labor-intensive, which means the people who receive the most of it have historically been those with access to smaller classes, private tutors, specialized coaches, attentive supervisors, or unusually generous institutions.
Generative AI is beginning to change that equation.
Signal
McKinsey and Junior Achievement Worldwide recently introduced JA Pitch Master, a multimodal generative-AI coach designed to help students practice communication.
A student gives a pitch. The system evaluates not only what was said, but how it was delivered and how effectively it engaged an audience. It currently assesses fifteen dimensions, with plans to expand that number significantly, and returns individualized feedback intended to help the student try again.
And again.
And again.
The important feature is not simply that artificial intelligence can critique a presentation. We already know machines can evaluate language.
The important feature is that the student gets something that has historically been difficult to provide at scale:
another rep.
There is no scheduling the next coaching session. No embarrassment about asking the same question twice. No exhausted instructor trying to provide meaningful feedback to the twenty-eighth student of the afternoon.
The practice space is effectively unlimited.
McKinsey and Junior Achievement have also surrounded the experience with responsible-AI controls intended to moderate tone, detect bias, reduce hallucinations, and prevent feedback from becoming destructive rather than developmental.
The ambition is larger than better presentations.
It is to take a form of high-frequency coaching once concentrated in boardrooms, private schools, debate programs, and executive-development programs and make it available to a student with an internet connection.
That is an interesting use of AI.
Not because the machine is speaking for the student.
Because it is helping the student learn to speak.
Pattern
Much of the public conversation about generative AI has focused on the work the machine can perform.
Write the memo.
Summarize the article.
Analyze the spreadsheet.
Generate the image.
Prepare the presentation.
But the more routine cognitive production moves toward machines, the more valuable another category of capability becomes.
The ability to explain.
To persuade.
To listen.
To improvise.
To read a room.
To defend an argument.
To change your mind without surrendering your judgment.
To say what you actually mean when another human being is looking back at you.
We tend to bundle these together under the diminishing label of soft skills.
They are becoming anything but soft.
Communication is emerging as one of the primary interfaces between human judgment and machine capability.
The machine can generate twenty possible strategies. A person still has to decide which one deserves conviction.
The machine can prepare a flawless slide deck. A person still has to stand in front of a skeptical audience and make the case.
The machine can suggest the words for a difficult conversation. A nurse, manager, teacher, caregiver, entrepreneur, or parent still has to recognize what is happening in the other person while the conversation unfolds.
This is why the emerging evidence around AI-assisted practice deserves attention.
Students who use large language models as critics of argumentative writing can improve not only the drafts created with AI assistance, but subsequent work completed without it. Learners practicing difficult conversations with synthetic patients or avatars can improve performance when those interactions later become real. Language learners can become more willing to speak when the cost of making a mistake falls.
The machine creates a place to rehearse.
And rehearsal matters.
But there is another pattern running beside it.
Students are also increasingly using the same systems to avoid the cognitive work entirely.
Generate the answer.
Write the essay.
Solve the problem.
Complete the assignment.
The same technology capable of strengthening a skill can remove the need to exercise it.
That apparent contradiction is not really a contradiction at all.
It is a design problem.
Implication
The central question is therefore not whether students should use artificial intelligence.
They already do.
The more useful question is:
What should the AI be designed to make the human do?
That distinction changes everything.
A system designed primarily to produce an answer encourages substitution.
A system designed to critique an answer encourages iteration.
A system that writes the speech can reduce practice.
A system that listens to the speech can multiply it.
A system that resolves every difficulty can weaken agency.
A system that introduces productive friction can strengthen it.
The difference is subtle in interface design and enormous in human consequence.
JA Pitch Master is interesting because its basic architecture points in the second direction.
The human performs.
The machine observes.
The machine responds.
The human interprets the response.
The human performs again.
That is a practice loop.
And generative AI may make extraordinarily dense practice loops economically possible.
For education, that has important implications for equity.
Historically, individualized feedback has been scarce because human attention is scarce.
Affluent families could purchase more of it.
Well-resourced schools could provide more of it.
High-potential employees could be selected for executive coaching.
Everyone else received whatever attention the institution had left.
AI changes the marginal cost of another rehearsal.
A student in rural Vermont, an under-resourced urban school, or a community workforce program can theoretically receive hundreds of rounds of structured feedback without requiring hundreds of hours of additional instructor capacity.
That does not eliminate inequality.
Access to devices, connectivity, good teachers, stable environments, and social capital still matters enormously.
But it changes one important variable:
practice itself can become abundant.
That possibility extends far beyond school.
Imagine a direct-care worker rehearsing a difficult conversation with a family member before walking into the room.
A first-time supervisor practicing corrective feedback before giving it to an employee.
A teenager preparing for a first job interview.
An entrepreneur rehearsing a pitch.
A nursing student practicing how to explain a frightening diagnosis.
A community leader preparing to speak at a public meeting.
A new caregiver learning not only what to say, but how to listen.
The value is not that the artificial conversation replaces the real one.
Its value is that the artificial conversation prepares the person for the real one.
Action
If this is the opportunity, then we should design for repetition with transfer.
Start with the practice loop.
Ask what human capability should be stronger after the machine disappears.
Then build backward from that outcome.
A writing tool might require the student to defend why a suggested revision should be accepted or rejected.
A presentation coach might ask for three consecutive attempts rather than producing a polished script.
A workforce-training platform might score progress in listening, clarity, empathy, adaptability, and confidence, then verify those gains through a live interaction.
An educational program might deliberately alternate between assisted and unaided work so that residual human capability remains visible.
The goal should not be perfect AI-assisted performance.
It should be better unassisted humans.
This also suggests a different approach to AI literacy.
Prompt engineering is useful, but the deeper competence is learning how to operate inside an iterative relationship with a machine:
state an intention,
set constraints,
evaluate the response,
identify what is wrong,
refine the approach,
and retain responsibility for the final judgment.
That is not merely a technical skill.
It is a form of agency.
For schools, youth workforce programs, caregiver academies, community-development initiatives, and employers preparing people for an AI-saturated economy, confident communication should therefore become a measurable outcome rather than an assumed byproduct.
Give people more chances to speak.
More chances to fail safely.
More chances to hear themselves.
More chances to adjust.
More chances to discover that what sounded clear in their head did not sound clear in the room.
There is no reason to accept the old scarcity of practice simply because we are accustomed to it.
Undersong
We have spent much of the first generative-AI era asking what happens when machines become capable of producing human language.
There is another possibility hiding underneath that question.
What happens when machines make it possible for more humans to find their own?
A machine can now listen without becoming impatient.
It can respond without embarrassment or ego.
It can sit through the tenth attempt as readily as the first.
That is not a replacement for a teacher, coach, mentor, colleague, audience, or friend.
It is something quieter and potentially more consequential:
a rehearsal space.
And perhaps one of the most valuable uses of artificial intelligence will not be helping us avoid difficult human performances.
It will be giving more people enough practice to enter them.
Because when the slides disappear, the prompt window closes, the interview begins, the family sits down, the patient looks up, or the room goes quiet, the technology reaches a boundary.
Someone still has to speak.
And the voice that carries weight will belong to the person who has practiced enough times to know what they mean.