According to Business Insider, Blackstone CTO John Stecher says there’s a definitive shift in entry-level software engineering. He graduated in 2001 when only about 260 students were in computer science at his alma mater, the University of Wisconsin-Madison; by fall 2025, that number is projected to be roughly 2,500. Nationally, computer science bachelor’s degrees conferred in 2021-2022 were up 115% from two decades ago. Stecher says today’s new engineers have “insane skill sets” that far surpass his own starting out, but they have fewer chances to learn core skills on the job because AI tools now automate the basic tasks that once served as training. To combat this, Blackstone has implemented a “buddy system” pairing new hires with senior colleagues to help them learn when AI is right and when it’s wrong.
The Talent Paradox
Here’s the thing: Stecher is pointing out a fascinating paradox. The raw, upfront talent entering the field is, by his own admission, astronomical. These folks aren’t just reading books; they’re leveraging Google, ChatGPT, Claude, and entire ecosystems of instant knowledge from day one. That’s powerful. But it creates a weird gap. They can *generate* code, but have they built the foundational intuition for *why* that code works, or when it’s a brittle solution? Probably not. And that’s a problem no one really saw coming a decade ago. We used to worry about engineers not knowing how to code. Now we have to worry about engineers not knowing how to *think* about coding, because the machine does the first ten drafts.
The Apprenticeship Crisis
Stecher’s comment about functional tests hits the nail on the head. In the past, grunt work like writing tests or debugging a narrow module was the golden apprenticeship. It forced you to understand a system intimately, to see how the gears fit together. Now, as he says, “A lot of the AI tooling can do that for you.” So what’s left for the new person? The high-level, abstract problems? You can’t start there. It’s like trying to teach someone architecture by having them design a skyscraper on day one, without ever letting them lay a brick. The “buddy system” is a decent, human-centric response. But let’s be skeptical: in a high-pressure corporate environment, how often will that “buddy” truly have the time to mentor, versus just answering Slack questions between meetings? The structural learning gap created by AI might be bigger than any buddy program can fix.
The Shifting Skill Set
This is where the real change is happening. The job is becoming less about syntax memorization and more about being a brilliant editor, critic, and systems conductor. You need to break down logical problems, know which tool to ask, and—critically—question the output. That last part is everything. The most dangerous engineer in the world today is the one who blindly trusts the AI’s hallucinated code. So Stecher is right, that discernment is the new core skill. But how do you teach gut instinct and critical thinking? You can’t exactly prompt-engineer that. It comes from making mistakes, from seeing systems fail, from the very hands-on experience that’s being automated away. It’s a vicious cycle.
A Broader Trend With Deeper Roots
Look, this isn’t just about AI. It’s the culmination of a long trend towards abstraction. Frameworks abstracted away server management. Cloud abstracted away hardware. Now AI is abstracting away basic logic. Each step makes us more powerful but also more distant from the underlying machinery. And while the number of CS grads has exploded, some programs are now seeing declines as the hype cools and job security fears rise. So we might be heading for a double squeeze: fewer students entering the pipeline, and those who do enter facing a career where the foundational learning phase is murkier than ever. Companies that figure out how to rebuild that apprenticeship model—not just with buddies, but with structured, intentional practice—will have a huge advantage. Everyone else will just have very talented engineers who are really good at asking questions, but not always great at knowing the answers are wrong.
