Every generation of developers has been told their jobs are about to disappear.

Compilers were supposed to eliminate programmers. Virtualization was supposed to eliminate sysadmins. Cloud was supposed to eliminate data centers. Automation was supposed to eliminate ops teams.

None of it happened.

What actually happened: expertise relocated. It did not disappear. Punch card operators became systems programmers. Sysadmins evolved into infrastructure engineers. Ops teams became DevOps, then platform engineering. Data center workers? Still racking servers in those AWS facilities. The ones who wanted to shift moved into cloud architecture.

The pattern is not replacement. It is lift and shift.

The Vibe Coding Debate

We are in another transition now. AI-assisted development is real, and the term “vibe coding” has entered the conversation. The framing matters for how leadership responds.

In February 2025, Andrej Karpathy, co-founder of OpenAI and former Tesla AI director, coined the term in a tweet that went viral. He described it as “fully giving in to the vibes, embracing exponentials, and forgetting that the code even exists.”

The term has become a tribal marker. One interpretation says engineers are becoming obsolete, so reduce headcount and let AI generate the code. The other interpretation recognizes this as the next abstraction layer, requiring judgment and experience to deploy effectively, just like every previous layer.

The data supports the second interpretation.

What the data actually shows

Y Combinator reports that 25% of their Winter 2025 cohort has codebases that are 95% AI-generated. The headline sounds like replacement. The detail tells a different story.

“Every one of these people is highly technical, completely capable of building their own products from scratch. A year ago, they would have built their product from scratch. But now 95% of it is built by an AI.”

That is YC managing partner Jared Friedman explaining the statistic. The founders using AI most heavily are not less technical. They are more productive.

Andrew Ng, former head of Google Brain and Stanford AI professor, has pushed back on the casual framing. He calls AI-assisted coding “a deeply intellectual exercise” and warns that the term “vibe coding” misleads people into thinking you just “go with the vibes.”

Simon Willison, a respected developer and creator of Datasette, draws a critical distinction: “If an LLM wrote every line of your code, but you’ve reviewed, tested, and understood it all, that’s not vibe coding in my book. That’s using an LLM as a typing assistant.”

The mentorship vacuum

The vibe coding debate has a blind spot: who is mentoring the next generation?

The traditional model was apprenticeship through grunt work. You learned by reading existing code to understand patterns. You wrote boilerplate that forced you to internalize system shapes. You made mistakes in low-stakes contexts where seniors caught them. You debugged your own broken code, which taught you how things fail.

AI short circuits several of these feedback loops simultaneously. If you can generate working code without understanding it, you skip the struggle that builds mental models.

But here is what the doomsayers miss. We have been here before and creativity did not die. It adapted.

The Pattern of Knowledge Redistribution

When compilers replaced hand coded assembly, engineers did not stop learning how systems worked. They started sharing patterns, building libraries, teaching each other the new abstractions. The initial machine generated output became a starting point for iteration, not an endpoint.

When Stack Overflow emerged, critics said developers would stop understanding code. They would just copy paste answers. Instead an entire ecosystem of knowledge exchange evolved. People learned by seeing how others solved problems then improving on those solutions.

The pattern is not knowledge atrophy. It is knowledge redistribution. Early iterations get shared. Communities form around making them better. The humans who understand why something works teach the humans who only know that it works. Creativity does not disappear. It migrates to higher level problems.

The Two Camps Raising the Alarm

That said the migration is not automatic. It requires intentional mentorship. And that is where I see a vacuum forming.

The people raising this concern fall into two camps.

Active Mentors

They see the gap in real time. Junior devs who can prompt their way to a working feature but cannot explain why it works. Who freeze when AI generated code breaks in production. Who lack the failure library that experienced engineers accumulate over years.

Reluctant Mentors

They know they should be teaching but are not. The pressure to ship has not decreased. If anything AI has raised output expectations. Mentorship takes time that does not show up in sprint velocity.

Both camps are pointing at the same vacuum. Neither has a clear solution.

What Different Collaboration Looks Like

We are running an uncontrolled experiment on an entire generation of engineers. Seniors learned in a pre-LLM environment. They are mentoring juniors into a post-LLM environment. The skills that made seniors successful may not be the skills juniors need.

But history suggests the path forward is not less collaboration. It is different collaboration.

Code review evolves from “did you write this correctly” to “do you understand why this works.” It also checks “when it will not.”

Mentorship shifts from teaching syntax to teaching judgment. Pattern recognition. Failure modes. Architectural tradeoffs.

Communities form around prompt strategies and model limitations the same way they formed around languages and frameworks.

Juniors who cannot yet evaluate code quality learn by watching seniors reject, modify, and improve AI output.

Strategic Implications for Leadership

The companies that figure this out will have a meaningful advantage in five years. The ones that assume AI solves the talent pipeline will discover they hollowed out their engineering bench.

Productivity Gains Require Experienced Judgment

AI accelerates output. It does not replace architectural thinking, security awareness, or system design. Teams without senior engineers reviewing AI-generated code are accumulating technical debt faster than they realize.

The Junior Developer Pipeline Is Changing, Not Ending

Entry-level engineers will learn differently than previous generations. The mentorship model needs to evolve. But the need for people who understand systems remains.

The Winners Will Treat This as a Skill Transition

The pattern from punch cards to compilers to cloud is consistent: companies that invested in their people captured the gains. Companies that treated new tools as a reason to cut headcount lost institutional knowledge they could not rebuild.

A Different Starting Point

I work with veterans transitioning into tech through Operation Code. They bring something most bootcamps do not teach. Operational thinking and pattern recognition developed before anyone touches a keyboard. They understand that accomplishing the mission is the point, not the specific tool you use to get there.

That mindset might be closer to what the next generation needs than traditional CS curricula.

Every previous abstraction layer created new ways for people to teach each other. This one will too. The question is whether we are intentional about building those pathways or just hope they emerge.

The Bottom Line

The abstraction layer is shifting. The need for human judgment is not going anywhere.

Claiming that AI will eliminate developers because we no longer write assembly is misguided. It’s like suggesting we cannot build compilers because we no longer use punch cards. The skill evolves. It does not disappear.

The organizations that understand this will invest in their people, evolve their mentorship models, and capture the productivity gains. The ones that see AI as a headcount reduction strategy will discover what every previous generation learned the hard way. Abstraction relocates expertise. It does not eliminate it.

Sources

Karpathy, Andrej. “Vibe coding” tweet. February 2025. x.com/karpathy/status/1886192184808149383

Ng, Andrew. Remarks at LangChain Interrupt conference. Reported by Business Insider, June 2025.

Willison, Simon. “Not all AI-assisted programming is vibe coding.” March 2025. simonwillison.net/2025/Mar/19/vibe-coding

Y Combinator W25 statistics. TechCrunch, March 2025. techcrunch.com