There is a headline going around this week that everyone wants to be true.
AI was supposed to gut engineering. Instead, new data from SignalFire says engineering is the most resilient function in tech. At the big tech companies, total hiring is down 25% against the 2019 baseline, but engineering hiring is down only 11%. Engineers are now 55% of all hires, up from 46% in 2019. Jensen Huang says software engineers are "busier than ever."
If you write code for a living, you read that and exhale.
Don't exhale yet.
The case is real
The resilience is genuine, and worth defending before I take the headline apart.
The numbers come from a firm tracking millions of people across tens of millions of companies, and they line up with what most teams feel. Demand for people who can ship software has not collapsed. If anything, the better the tools get, the more software everyone wants. That is just Jevons paradox wearing a hoodie: make a resource cheaper to use and total consumption goes up, not down. Cheaper code means more code, more products, more things that need someone who understands them.
So yes. The senior engineer with a good agent is more valuable this year than last year. That part is true.
It is also not the part of the report that matters.
The footnote nobody is quoting
Here is the same dataset, the line that did not make the headline.
New-graduate hiring at the big tech companies is down more than 50% since 2019. Fresh grads are now about 7% of all hires. The average age of a technical hire has gone up by three years since 2021, because companies have quietly stopped being willing to train anyone.
Read those two stories together and the cheerful headline falls apart.
"Engineering is resilient" is a senior's headline. The resilience is concentrated almost entirely in people who were already established. The first rung of the ladder, the one every current senior climbed, has been sawn off.
That is a job market drawing down a stock it has stopped replacing. Call it whatever you like, but not healthy.
Why every quarter votes to do this
The maddening thing is that no villain is required. Each individual decision is rational.
A senior engineer with an AI agent out-produces a senior engineer plus a junior, this quarter, on this roadmap. The junior is slower, needs review, asks questions, breaks things, and takes a year to pay back the investment. The agent is available now and bills by the token. So when budgets tighten, the junior is the line item that gets cut, and the work still gets done.
Run that logic across an entire industry and you get exactly the SignalFire numbers. Nobody decided to stop making engineers. Everybody just decided, separately, that this junior, on this team, was optional.
Rational per quarter. Suicidal per decade.
What "resilient" is actually measuring
Look closely at which engineers are thriving and you notice they all have the same thing. Agents took raw coding speed off the table for everyone. What is left, and what the survivors trade on, is judgement: knowing what to build, spotting the bug the model confidently introduced, sensing when an abstraction is going to rot. I have written before that speed got cheap and judgement didn't, and the hiring data is that argument showing up on a balance sheet.
But judgement is not downloaded. It is the residue of years of doing the unglamorous work: shipping the small feature, owning the on-call page at 3am, watching your own clever code become next year's mess. The junior years are the factory that produces the senior judgement the market now pays a premium for.
We have decided judgement is the scarce, valuable thing. And in the same breath, we have stopped funding the only process that makes it.
The seed corn
There is a reason a starving farmer who eats next year's seed corn is the oldest cautionary tale we have. The meal is real. The field is empty in spring.
The engineering pipeline has a five-to-ten year lag. A grad hired today is a competent mid-level engineer in three or four years and a genuine senior in seven or eight. Which means the seniors who are "resilient" right now were planted in 2015, 2017, 2019, back when the industry still hired and trained people who didn't know anything yet.
The shortage we are creating won't show up on a dashboard in 2026. It shows up around 2031, as a generation of seniors retires or burns out and there is a thin, expensive, undertrained cohort behind them. By then the cause is years in the past and unfixable on any quarterly timeline.
Call that resilience if you want. It looks more like coasting on a full tank with the fuel pump switched off.
What to actually do
If you run a team: hire at least one junior you do not strictly need this quarter, and treat training them as building inventory, not charity. The agent multiplies people who already have judgement. It cannot manufacture the judgement itself, that was the pipeline's job. The cheapest seniors you will ever hire are the juniors you train now.
If you are early in your career: the on-ramp got narrower, not closed. The skill that compounds is not "can prompt an agent," everyone has that. It is the judgement the resilient seniors are being paid for. Get it the only way it is gotten: own real systems, read more code than you write, and refuse to ship anything you cannot explain. Be the junior who is obviously becoming a senior, because the market is short on exactly that and getting shorter.
The headline is right that engineering survived the AI scare. It just left out who paid for the survival, and when the invoice comes due.
Engineers are resilient. The profession is quietly eating the thing that makes more of them.