The Economics of Mentorship in and Beyond 2025

Who pays for mentorship in the age of AI... and why?

Introduction

My friend Navin asked a question at the end of his recent blog post:

The biggest unanswered question is:

The way we get seniors in an industry is by having juniors do the grunt work for years and slowly learn from the seniors until they become seniors themselves. But if the grunt work is being done by AI, where will the next generation of seniors come from? This is a serious problem across many industries, possibly the entire world economy, and I haven’t yet heard any convincing answers of how we are going to solve this.

Have you?

Leave aside, for now, the question of how we are going to solve this. Step 1 with any problem is understanding it in its entirety. So let’s try that with this problem.

(Note: I make clear at the end of this post that the idea is for you to poke holes in this toy model. Finding out ways this might be wrong is a good way to start understanding what we need to solve for in the time to come. So if you think this model is not sufficiently developed, great! But please do tell me why)

A Simple Toy Model

Let’s say that there is a fictitious world in which there exist Juniors (J), Seniors (S) and both of these creatures work in places called Firms (F).

1. Seniors can earn money by doing their actual work (let’s call this wages_work ).

2. But they can also earn money by mentoring juniors. Let’s assume that the payment for this kind of work is called wages_mentorship

3. Juniors get paid wages_juniorwork

4. The work that they do generates Value_juniorwork

5. Let’s further assume that juniors can only become seniors by being mentored¹.

6. Let’s say that the value of this mentorship, for J’s, is value_mentorship

Now, in the good ol’ days, the system worked because

Value_juniorwork > wages_juniorwork (1)

This could be because wages_juniorwork were very low, or it could be because the work that was done had positive externalities, or both.

TL;DR

Before AI, the system worked because:

• Firms hired juniors because (1) “worked”

• Juniors learned by proximity to seniors doing work

• Over time, juniors became seniors

• No explicit payment for mentorship was needed where senior workers were concerned—it was bundled with employment

Our Toy Model Meets AI

But in our brave new world? Well, in our brave new world, the fees “paid” to AI are even lower than the wages we pay Juniors (INR 2000 per month). AI works 24/7, has infinite patience and is getting better all the time. And because:

wages_juniorwork > fees_AI. (2)

… Juniors find it increasingly difficult to find gainful employment. But per pt. 5, Juniors can’t become Seniors without mentorship. So the world needs to find a way to pay seniors to get them to mentor Juniors.

Now, it is a fair assumption that:

wages_work > wages_mentorship. (3)

Why is this a fair assumption? Well, Seniors can earn more doing actual work than mentoring. Why? Because AI has made them more productive at their work. The same technology that eliminated junior jobs made senior work more valuable. This is the cruelest irony, unfortunately: the tool that broke the ladder also made being at the top of the ladder more lucrative.

TL;DR:

AI changes two things simultaneously:

• Makes juniors unemployable per (2)

• Makes seniors more valuable, raising the opportunity cost of mentorship per (3)

• This unbundles mentorship from employment—now it must be an explicit, paid transaction

THKK?

(Toh Humanity Kya Karein? Sorry, couldn’t resist)

And so we need to figure out a way for Juniors to make up the difference in (3). Either Juniors pay for this themselves right now (through their own savings, their parents’ savings, by borrowing) or by pledging that Seniors get a part of the current Juniors’ future income streams. But even if they use their own or their parents’ savings, this only makes sense if the NPV of future earnings exceeds the investment.

That then begs the question, what is the present value of the future income stream of these Juniors?

\(V = \sum_{t=5}^{T} \frac{\mathit{wages}_{\text{senior},\, t}-\mathit{wages}_{\text{junior},\, t}}{(1+r)^{t}} \)

You shouldn’t be asking if it is worth putting in the effort to become a senior unless the wage differential is sufficiently high. Now, that much is obvious - why does this equation then look so very ghastly?

What this equation is saying is this:

1. Let’s assume that a Junior starts as a mentee, and is mentored for five years

2. After five years, the mentee becomes a senior

3. Then, from that point on, this person will be a Senior until they retire in year T

4. During this time (five years from now until they retire), they will earn not just the wages they would have if they were a junior, but also a wage premium for being a senior.

5. This equation simply discounts this “Senior premium”, earned over their entire career, to its Net Present Value today.

So mentorship only works if V is positive and sufficiently high over whatever horizon you want to think about². What is your best guess for what wages_senior,t will be t years from now? Will they be higher than the than wages being paid to juniors over these five years?

If AI is good enough to replace junior workers today, what are the chances it will be good enough to replace senior workers tomorrow, or five years from now? Nobody knows the answer for sure, of course, but what do your assumptions say? What if you are rather pessimistic about any human employment five, ten, fifteen or twenty years from now?

Or, to put it in plain simple English: under the most pessimistic scenario, there is no reason to invest in mentorship today, and there is already no reason to hire Juniors today.

Now, you can of course say that the most pessimistic scenario will not materialize. And therefore this model does not work. You can come up with twenty other reasons for why this model does not work. Each of those reasons implies that there are scenarios in which Junior unemployment does not fall today. It also implies that there are scenarios in which mentorship makes sense, and remains important.

And so poking holes in this model becomes Very, Very Urgent.

We’ll try and do that in the next blog post. But in the meantime, have at it yourself: tell me how my model does not make sense.

I and the Juniors will be most grateful!

1. No, this is not necessarily entirely true. But it has the magnificent benefit of being a simplifying assumption, so shush.

2. I’m assuming it takes 5 years to transition from a Junior to a Senior, but this can be any number you like