Will AI cause a senior engineer shortage? I don't buy it.
AI is genuinely eating parts of junior work. The leap from there to a coming senior shortage doesn't follow. The industry has watched whole categories of work get automated, outsourced, and abstracted away, and adapted every time.
tsukumo
Short version: AI is definitely replacing parts of junior work. That part is real, and pretending otherwise is silly. What I don't buy is the next move everyone makes, that this automatically leads to a shortage of senior talent. It assumes juniors become seniors by repeating the exact journey we took. They don't. I've been in this industry long enough to watch entire categories of work disappear, get automated, outsourced, or abstracted away. The field adapted every time. It will again.
The argument is tidy, which is part of why it spreads. It goes like this. AI is good at the small, well-specified tasks we used to hand to juniors. Juniors learned the craft by grinding through those tasks. So if AI does the grinding, juniors never accumulate the reps, never level up, and in ten years there's a hole where the senior bench should be.
I get why it's convincing. Every step sounds true on its own. And the first one mostly is.
The trouble is the middle step. The whole thing hinges on the idea that seniority is a fixed staircase, that you become senior by personally performing a specific sequence of tasks, in order, the way I did. Pull out the bottom stairs and nobody can climb.
That's not how it worked for me. It's not how it worked for anyone I came up with.
Here's the thing the panic always forgets. We have run this experiment. Repeatedly.
People used to hand-write assembly because there was no other way to get performance out of a machine. Then compilers got good, and a skill that had defined competent programmers for a generation became something almost nobody does by hand. Whole careers were built on register allocation. The compiler ate it.
We did not run out of good engineers. We got more of them, working on harder things.
Then manual memory management. For years, knowing how to not leak memory, when to free, who owns this pointer, was a core part of being trusted with real code. Garbage collection arrived and lifted most of that off most engineers. The hand-wringing at the time sounded familiar: kids these days won't understand what's really happening, they'll never learn rigor.
They learned plenty. They just learned it one layer up.
Categories the field abstracted away, then kept producing seniors
What disappeared
What replaced it
What happened to the engineers
Hand-written assembly
Optimizing compilers
Moved up to systems and applications
Manual memory management
Garbage collection
Built bigger systems, fewer crashes
Racking and patching servers
Cloud and infra-as-code
Became the people who design the cloud
The sysadmin who racked servers and ran cables didn't go extinct when the cloud showed up. The good ones became the people who architect what runs on the cloud. The job moved. It didn't end.
So how do juniors become senior if AI does the junior work?#
The same way they always have. By solving the problems of their generation.
This is the part the staircase model gets wrong. A junior engineer in 1999 did not become senior by mastering punch cards or hand-tuned assembly. That work was already gone. They became senior by wrestling with the problems that were actually hard in 1999, which were different from the ones hard in 1985, which were different again from now.
Nobody became a senior engineer by repeating the journey of the generation before them. You can't. The journey isn't available anymore. The tools have changed by the time you arrive.
What's available to a junior today is the stack built on top of AI. Orchestrating agents that write the first draft. Knowing when the plausible-looking output is quietly wrong. Designing the system, the verification, the gates around tools that are confident and frequently mistaken. That's not junior work that AI took. That's the new hard problem, and it's a senior-shaped one. The job is moving, not ending, and the people who figure out the moved version are exactly who becomes senior next.
Strip away the tooling and ask what a senior engineer actually has that a junior doesn't. It isn't a memorized API. It isn't the ability to write a sort by hand.
It's judgment under uncertainty. Knowing what to build and what to refuse to build. Smelling where a system is going to break before it does. Making a call when the spec is wrong, the data is incomplete, and someone needs an answer today. Knowing which of the ten plausible approaches is the one that won't haunt you in eighteen months.
None of that is a tool. None of it is a task AI can lift away, because it isn't a task at all. It's what's left after the tasks are automated. It's the residue of having solved real problems and lived with the consequences.
AI changes which problems are hard. It changes which tools you reach for first. It does not change what experience is made of, because experience was never the typing. It was the deciding.
I'm not claiming there's nothing to manage here. There is.
If you automate every entry-level task and then hire zero juniors because "AI does that now," you will, in fact, have a pipeline problem in a few years. That's not AI causing a senior shortage. That's a hiring decision causing one, and blaming the tool. The companies that keep bringing in juniors and point them at the new hard problems, the agent-shaped, system-shaped problems, will have plenty of seniors. The ones that quietly stop hiring early-career people will reap what they planted.
The mistake is treating juniors as cheap labor for tasks. They were never that. They were seniors who hadn't happened yet. Give them the current generation's problems instead of the last generation's busywork, augment them instead of replacing them, and the pipeline takes care of itself the way it always has.
We run agent fleets to ship our own software, so I'm not theorizing about this. I watch it weekly. The agent writes a lot of the code that used to be a junior's first month. And the work that's left, deciding what the system should do, catching the confident-but-wrong output, owning the result, has gotten more senior-shaped, not less. There's more output per person now. There's also more need for judgment, not less.
So no, I don't think AI is coming for the senior bench. I think it's coming for the part of the job that was never where seniority lived in the first place.
The tools change. The path changes. The problems change. Experience doesn't.
We help eng leaders separate what AI actually automates from what they still need humans for.
There's no strong reason to think so. The fear assumes that if AI does the junior tasks, no one accumulates the experience to become senior. But seniority was never about repeating a fixed set of tasks. The field has automated, outsourced, and abstracted away whole categories of work before, assembly, manual memory management, racking servers, and kept producing seniors. The path changes; the people still climb it.
If AI does junior work, how do juniors become senior?
The same way every generation did: by solving the hardest problems available to them, which are no longer the problems we solved. A junior in 1998 didn't earn seniority by writing assembly; that work was already gone. They earned it on the web stack of their day. Today's juniors won't become senior by hand-writing the code an agent now writes. They'll do it by owning the systems built on top of agents, the part that's still hard.
Does automating junior tasks break the engineering talent pipeline?
Only if you believe the pipeline is a fixed sequence of tasks you must personally perform in order. It isn't. Garbage collection removed manual memory management from most engineers' daily work, and the industry produced more capable engineers afterward, not fewer. Removing a layer of toil has historically freed people to work at a higher level sooner, not blocked them from advancing.
What does seniority in engineering actually come from?
Judgment under uncertainty. Knowing what to build, what to skip, where systems break, and how to make a call when the data is incomplete. None of that is a specific tool or task; it's what accumulates from solving real problems with real consequences. AI changes which problems are hard and which tools you reach for. It doesn't change the thing experience is actually made of.
