If Agents Fix the Small Bugs, Where Do Senior Engineers Come From?
The question I get asked most about AI in engineering has stopped being about tooling. It is some version of this: if agents now fix the small bugs and work the support tickets, where does the next generation of engineers come from?
It is a fair question. The first rung of the ladder used to be the simple bug fix, the test nobody wanted to write, and the CRUD endpoint. CRUD stands for create, read, update and delete: the routine plumbing that lets an application save a record, read it back, change it and remove it again. Every system needs an enormous amount of it, very little of it is interesting, and it went reliably to whoever had joined most recently. That work has largely gone to the machines, and graduate hiring appears to be following it down.
The conclusion most people draw is that the ladder is broken. I think that is the wrong reading, and the right one is less comfortable: the ladder was never the typing; it was the judgment, and the route to judgment was always accidental.
What the grunt work was actually for
Very few organisations ever hired a graduate because they needed more CRUD endpoints. They hired one because after three years of small fixes and support tickets that person had seen a few hundred things fail in small ways, absorbed the codebase, watched senior people argue about design, and started to develop a sense of what good looks like.
The tickets were the delivery mechanism; judgment was the payload. We got the payload free as a side effect of needing the work done, which meant nobody had to design the learning, because it happened while people were busy.
That side effect has now been removed, which does not mean judgment stopped mattering. It means we have to develop it on purpose.
Developing it on purpose
I have been writing about protected surfaces: the parts of an estate where human ownership stays mandatory no matter how good the tooling gets. Architecture, identity and access, contracts, customer data, and the quality gates themselves. Agents propose there; people decide.
Those surfaces are also the best training ground anyone has, because they are exactly where the judgment lives. So the deliberate version of the ladder looks something like this.
- Rotation under named owners. A junior spends time on each protected surface, working alongside the senior engineer who owns it. Apprenticeship rather than osmosis. The senior owner is accountable for the surface, and part of that accountability is developing the next person who could own it.
- Spec authorship before implementation. Writing down what correct behaviour looks like, before any code exists, is judgment training in close to its purest form. It forces you to think about edge cases, failure modes and what the business actually meant. It also happens to be the work that keeps agents honest, so it is real output rather than an exercise.
- Escalation exposure. Put juniors where the failures surface: the review tier, the incidents, the things agents got confidently wrong. Very little teaches faster than watching something plausible turn out to be false.
None of that is charity. Every item on that list is work the organisation needs done anyway.
Where the ladder now leads
The other half of the question is what the top of the ladder looks like now, and I think it has changed shape.
For twenty years the reliable career was depth: becoming the person who knows JavaScript, or Kafka, or the payments stack better than anyone else in the building. That was rational while implementation was the expensive part, because deep specialists were how expensive things got built.
Implementation is no longer the expensive part. What costs now is knowing what to build, whether the design fits the estate, what must not be touched, and whether what came back is actually right. Those decisions do not sit inside a single stack; they sit across feature intent, architecture and implementation at once.
So the valuable engineer is the one who can hold all three in one head and direct agents across them. I have been calling this the full-breadth engineer, and it is not the same thing as a generalist who knows a little of everything. It is someone with enough depth in several places to have real opinions, and enough breadth to see how a decision in one place lands in another.
The obvious objection is that breadth becomes shallowness, and that you cannot have judgment about architecture you have never built. That is fair, and the answer is that depth still gets built, deliberately rather than accidentally. You own a protected surface for a period, you go deep on it accountably with a senior owner beside you, and then you rotate. The depth is sequenced instead of stumbled into, and the breadth accumulates across the sequence.
I think that is a better career than the one it replaces. The old path made you excellent at one thing and hoped the market still wanted that thing in fifteen years; this one makes you the person who decides, which has rarely gone out of demand.
The organisations that work this out early will have senior engineers in five years. The ones still waiting for juniors to learn by osmosis probably will not.