I keep getting the same question from other leaders:
senior engineers come from?
For years we called grunt work a cost - the boring tickets, the unreasonably difficult debug, reading someone's bad code until you understood why it was bad.
It was the curriculum.
We got the next generation "for free" as a side effect of the junior-level grind. Now we've automated those boring tasks ... and the learning went with it.
The big question is: what is the curriculum now?
It isn't reviewing AI output. Reviewing is recognition: does this look right? Approve. Recognition doesn't make you good at anything.
Aviation learned this the hard way. Pilots who only babysat the autopilot lost the ability to fly - and eventually they couldn't notice when the autopilot was failing.
A junior who only approves AI output
is that pilot.
So where does judgment actually come from now? Three places.
"But isn't owning the outcome just reviewing it with extra steps? You're still not writing the code."
Fair. The difference isn't just attitude. It's the work. It's the effort.
Rubber-stamping stops at "looks fine." Owning doesn't. You decided what to build, and you determined where it would break before it ran. You're also the one messaged at 9pm when it actually breaks, and then the one defending a decision when others push back.
Deciding, predicting, debugging, defending. That's where judgment gets built. Reviewing skips all of it.
But owning outcomes isn't enough on its own. You still have to put in the reps - work on real problems, with AI off, on purpose. Pilots still hand-fly for a reason.
Yes, it's slower. That's not waste. That's the tuition.
Mentorship can't stay a favour people squeeze in on their own time.
Medicine solved this a century ago: residencies, real supervision, counted as the job. Nobody expects an attending physician to teach residents "when things are quiet."
We still run grand rounds within our teams - senior engineers walking through real failures, real decisions, real trade-offs, in front of the people who need to absorb them. It works.
We've just created the most patient tutor a junior will ever have. It never gets tired of questions. It never makes anyone feel small for asking.
But that only pays off if you intentionally point AI at teaching, not just doing. The same model that writes the code can explain why the code is shaped that way, what the alternatives were, and what would break under load.
One produces output. The other produces engineers.
The curriculum has to become culture - built on purpose, by everyone, at a cost. Skip it for a faster quarter and you'll come up short of seniors exactly when you need them.
the scarce resource.
The people you can hand the
hard problems to will be.