The short version: The Stanford AI Index shows that writing code is no longer scarce: AI performance on SWE-bench Verified jumped from roughly 60% to near the human baseline in one year, and four out of five university students already use generative AI. A separate Stanford Digital Economy Lab study found entry-level software engineering employment fell close to 20% between late 2022 and mid-2025, while experienced engineers held steady. The value of an engineer is moving from execution to judgment, ownership, communication, and business acumen.
The new Stanford AI Index report just landed, and if you are early in your software career, you should read the workforce section twice. Pair it with a second Stanford study from the same year and the picture gets sharp: the bottom rung of the ladder is being rebuilt while you stand on it.
Here are the four things that matter most, what they actually say, and what to do before the change reaches you.
Prefer the 90-second version? Watch it here: youtube.com/shorts/uRj-yHYheMY
Key Findings at a Glance
- Coding is no longer scarce. On SWE-bench Verified, AI performance rose from about 60% to near the human baseline within a single year, and four out of five university students use generative AI in their work (Stanford AI Index).
- Entry-level software roles are shrinking. Stanford's Digital Economy Lab found a 13% relative employment drop for workers aged 22 to 25 in AI-exposed jobs, and close to 20% for entry-level software engineering between late 2022 and mid-2025, while experienced engineers held steady or grew.
- Junior roles are being redefined, not just cut. IBM is tripling US entry-level hiring, but the new job descriptions replace routine coding with judgment, oversight of AI output, customer interaction, and business acumen.
- Workforce reductions are expected. McKinsey's 2025 State of AI survey found about 30% of organizations expect to reduce workforce size within a year due to AI, with software and data engineering among the most affected functions.
For a deeper look at which skills actually protect your career, see the skills beyond code that determine who advances.
1. Writing code stopped being the thing that sets you apart
Four out of five university students now use generative AI in their work, according to the AI Index. On the coding benchmark SWE-bench Verified, model performance jumped from roughly 60% to near the human baseline in a single year.
Read that as one sentence: the ability to produce working code is no longer scarce. When a tool in everyone's editor can turn a clear spec into a passing function, the function is not where your value lives anymore. Knowing what to build and why is.
2. The entry-level squeeze is real, and software is at the front of it
This is the number that gets misquoted, so here is the accurate version. A separate Stanford study, from the Digital Economy Lab, tracked payroll data across millions of workers. It found a 13% relative drop in employment for workers aged 22 to 25 in the jobs most exposed to AI. For entry-level software engineering specifically, the decline was close to 20% between late 2022 and mid-2025.
Meanwhile employment for more experienced engineers in the same field held steady or grew.
Why the split? AI is absorbing the routine work that juniors used to cut their teeth on. The bug tickets, the boilerplate, the small well-scoped tasks. That work was never just output. It was how you proved yourself and earned the next level of trust. When it disappears, the traditional way in disappears with it.
3. The companies still hiring juniors have redrawn the job
Look at what IBM did. They are tripling US entry-level hiring, but the job they are hiring for is not the one from five years ago. The new descriptions cut routine coding and lean into judgment, oversight of AI output, customer interaction, and business acumen.
That is the tell. The companies that still want juniors are not asking “can you code.” They are asking “can you decide what is worth coding, catch where the AI got it wrong, and explain the tradeoff to a stakeholder.” The entry barrier moved from execution to judgment.
4. Workforce cuts are coming, and engineering feels the most pressure
McKinsey's 2025 State of AI survey found that around 30% of organizations expect to reduce workforce size in the next year as a result of AI. Software and data engineering sit among the most affected functions.
Leadership has stopped asking whether AI can do the work. They are now asking how many people they still need once it does. That question does not have a comfortable answer for anyone whose only contribution is code.
So here is the new reality
Juniors are squeezed. People who only code are the easiest to replace. People with solid skills beyond code are the ones who win.
When AI controls the how, your value moves to the why and the what. Ownership, communication, judgment, business acumen. Those stop being “nice to have” soft skills and become the actual currency of the job.
What to actually do about it
You cannot out-type a model, so stop competing on that. Compete where it is weak.
Take ownership of outcomes, not tickets. Be the person who asks what problem we are really solving before writing a line. That instinct is rare and it is exactly what survives.
Get good at explaining technical decisions to non-technical people. The engineer who can make a product manager understand a tradeoff is worth more than the one who silently ships the tradeoff.
Build judgment by reviewing AI output critically instead of accepting it. Every time you catch where the model was confidently wrong, you are training the one skill the report says leadership is now paying for.
Learn enough of the business to know which work matters. “Knowing what to build” is only possible if you understand who it is for and what it earns.
None of this means coding is worthless. It means coding alone is no longer enough to be safe. The developers pulling ahead right now are not the fastest typists. They are the ones who paired their technical skill with the human skills the machine cannot fake.
The ladder is being rebuilt. You can either wait to see what rung is left, or start building the skills that put you above the squeeze.
Utterskills trains the skills beyond code that this report says now decide who advances. If the data above made you reconsider where your time goes, that is the point.