Here is an uncomfortable truth that emerged this year: AI coding tools are everywhere and they might be changing software development in ways we didn't anticipate. Stack Overflow's 2025 Developer Survey shows 65% of developers now use AI tools at least weekly. That sounds like a success story. Until you notice that employment among software developers aged 22 to 25 fell nearly 20% between 2022 and 2025?precisely when these tools exploded in popularity.
The paradox deepens. At the same time adoption soars, trust plummets. Just 33% of developers trust the accuracy of AI-generated code, while 46% actively distrust it. Positive sentiment dropped from over 70% in 2023-2024 to 60% this year. Something is shifting and it is not just about better tools versus worse tools. It is about how we work, what we know and what happens when convenience replaces comprehension.
The "Vibe Coding" Warning Nobody Expected
The most striking critique of AI coding tools came from an unexpected source: Michael Truell, CEO of Cursor, one of the leading AI coding assistants. At Fortune's Brainstorm AI conference, Truell coined the term "vibe coding"?a method where developers close their eyes, ask AI to build something and accept the output without understanding what is happening under the hood.
His warning was blunt. "If you close your eyes and you don't look at the code and you have AIs build things with shaky foundations, as you add another floor and another floor and another floor, things start to kind of crumble." He likened it to constructing a house without knowing what is happening with the wiring or the foundation.
This matters because Cursor just closed a funding round at a $29.3 billion post-money valuation. When the CEO of a company selling AI coding tools warns against over-reliance on those very tools, that is not marketing spin. That is someone who sees the long-term consequences of short-term convenience.
What the Employment Data Actually Shows
Stanford University researchers used ADP payroll data to track developer employment trends. What they found is specific and troubling. Developers aged 22 to 25 saw employment drop 20% from late 2022 to 2025. Developers over 26? Stable or growing employment.
The timing is not subtle. The decline began in late 2022, right when ChatGPT and other generative AI tools launched publicly. Entry-level tech hiring fell 25% year-over-year in 2024. At the 15 biggest tech firms, entry-level hiring dropped 25% from 2023 to 2024.
Meanwhile, overall U.S. programmer employment fell 27.5% between 2023 and 2025, according to Bureau of Labor Statistics data. Software developers?a more design-oriented role?only declined 0.3% in the same period. The pattern suggests that routine coding tasks are being automated or absorbed, while strategic development work remains relatively intact.
This is not a temporary adjustment. This is a structural shift in what entry-level development work looks like?and whether it exists at all.
The Trust Problem Beneath the Surface
Adoption numbers tell one story. Trust numbers tell another. The 2025 Stack Overflow survey reveals that while 84% of developers use or plan to use AI tools?up from 76% last year?confidence in those tools is eroding.
Only 33% of developers trust AI-generated code accuracy. Nearly half actively distrust it. That gap between usage and trust is not sustainable. It suggests developers feel compelled to use tools they do not fully believe in, either because competitors are using them, because management expects speed gains, or because the baseline for "good enough" code has shifted.
The productivity gains everyone touts? They might be illusory. MIT Technology Review notes that "a growing body of research suggests that claimed productivity gains from AI coding tools may be illusory, with some developers on the front lines seeing initial enthusiasm wane as they bump up against the technology's limitations."
What looks like faster development in month one can turn into technical debt in month six. Code that ships quickly but breaks quietly. Systems that work until they don't and when they fail, nobody fully understands why because the person who wrote the code?the AI?isn't around to explain its reasoning.
What Changed and What Stayed the Same
AI coding tools excel at specific, bounded tasks. Need a boilerplate function? Generate it. Want to refactor repetitive code? AI handles that well. Looking for syntax you can't quite remember? Faster than Stack Overflow.
The problem emerges when these tools move from assistance to authorship. When developers treat AI as a coworker who writes entire features rather than a reference tool that speeds up known patterns. When "make it work" replaces "understand how it works."
Here is the uncomfortable part: junior developers are often the ones asked to maintain AI-generated code. They are the ones debugging systems they didn't write and don't understand, built on foundations they can't see. That is not a learning environment. That is a maintenance trap.
Senior developers with deep system knowledge can evaluate AI suggestions, reject bad ones and integrate good ones thoughtfully. Junior developers?the ones losing jobs at a 20% clip?are still building that judgment. They need repetition, mistakes and pattern recognition that comes from writing code, not just reading it.
The Real Cost Nobody's Calculating
Software development is not just about producing code. It is about building mental models of how systems work, why they fail and how they scale. Every line you write teaches you something about trade-offs, about edge cases, about what breaks under load.
AI tools short-circuit that learning loop. You get working code without the struggle that builds competence. The Stack Overflow survey shows 51% of professional developers use AI tools daily. That is half the profession potentially offloading the parts of development that teach you to become better at development.
The long-term question is not whether AI can write code. It demonstrably can. The question is whether a generation of developers who rely on AI from day one will develop the deep systems knowledge needed to architect complex software, debug subtle failures, or make sound technical decisions under pressure.
Cursor's $29.3 billion valuation suggests investors believe AI coding tools are the future. The 20% drop in young developer employment suggests that future might not include as many junior developers. Those two things are connected.
What This Means for Developers in 2026
If you are early in your career, the message is clear: do not let AI tools do the learning for you. Use them to speed up tasks you already understand. Resist the urge to copy-paste AI solutions without comprehending why they work. The developers who survive this shift will be the ones who understand systems deeply enough to evaluate AI output critically.
If you are hiring, consider what skills junior developers actually need now. It is not just writing code. It is reading AI-generated code, understanding its limitations and knowing when to reject it. That is a different skill set than what we have historically trained for, but it is the one the market is demanding.
If you are managing teams, Truell's warning about shaky foundations applies directly. Shipping fast matters. Shipping code nobody understands is a time bomb. Insist on code review. Insist on understanding. Treat AI as a tool that accelerates work, not a replacement for engineering discipline.
The industry is sorting into two groups: developers who use AI to amplify their capabilities and developers who use AI as a crutch. The employment data suggests the second group is struggling to find work. That is not a moral judgment. It is a market signal.
The Honest Assessment
AI coding tools are not going away. They will get better. They will write more complex code with fewer errors. That is the trajectory and resisting it is pointless.
But the idea that they are purely productivity multipliers?that they make every developer faster without tradeoffs?is not supported by the evidence. Trust is declining. Entry-level employment is collapsing. The CEO of a leading AI coding company is warning about foundational problems.
The developers who thrive in 2026 will be the ones who treat AI tools as exactly what they are: powerful assistants that require supervision. The ones who lose ground will be those who mistake convenience for competence, speed for understanding and code that compiles for code that works.
Vibe coding is real. And if Michael Truell is right, the foundations are already starting to crumble.
