The Rise of 'Vibe Coding': Are Junior Developers Obsolete?

[RightAbout24]

On February 2, 2025, Andrej Karpathy, the former Tesla AI director and OpenAI founding member, posted a short message on X describing what he called a new kind of coding, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. The post crossed 4.5 million views, spawned its own memes and counter memes, and within weeks the term vibe coding had appeared in the New York Times, the Guardian and Ars Technica. By the end of the year Collins Dictionary named it its Word of the Year for 2025. Eighteen months on, that single tweet has become the shorthand for a genuine and still unresolved argument running through the entire software industry, whether the entry-level developer job is quietly disappearing, and if so, what replaces it.

What vibe coding actually meant, and what it became

Karpathy's original definition was specific and, by his own account, fairly reckless. He described using tools like Cursor Composer, often paired with voice transcription software to minimize typing, accepting AI generated code changes without meticulously reviewing the diffs, pasting error messages straight back to the model for a fix, and letting the codebase grow organically, potentially beyond his own complete understanding of it. It was, in his framing, explicitly not meant for production systems, more a description of how to build fun throwaway projects, demos and explorations quickly. That precise definition did not survive contact with the wider industry. Within months, vibe coding had broadened into a catch-all term for any AI assisted programming, prompting comparisons to how the word agile lost its original precise meaning as it spread. The drift toward using it for production grade systems took on a genuinely negative connotation as teams ran into the predictable consequences of accepting code nobody had actually read.

Karpathy himself has since tried to draw a new line. In a follow-up post roughly a year later, in February 2026, he argued that vibe coding was already passe, replaced among professionals by what he termed agentic engineering, describing programming via LLM agents as increasingly becoming a default workflow for professionals, except with considerably more oversight and scrutiny than his original throwaway definition ever called for. In his own words, at the time of his original post, LLM capability was low enough that vibe coding mostly worked for fun projects and almost worked more broadly, the newer framing assumes far more capable models but reintroduces exactly the human review step the original term had explicitly discarded.

What the hiring data actually shows

The economic anxiety underneath this debate is grounded in some genuinely stark numbers, though the full picture is more contested than the scariest headlines suggest. Multiple industry analyses tracking US job boards found entry-level software engineering postings dropped roughly 67 percent between 2022 and 2024. Stanford's Digital Economy Lab, working with ADP payroll records covering millions of actual employees rather than just job postings, found employment specifically for developers aged 22 to 25 fell nearly 20 percent from its late 2022 peak, with the decline continuing into 2026. A separate industry analysis found the share of juniors and new graduates among total IT hires dropped from around 15 percent to approximately 7 percent over three years, meaning for every 100 new technology hires today, roughly seven are juniors where fifteen used to be. McKinsey research covering AI-exposed fields more broadly found a 16 percent relative decline in early-career employment specifically, while employment for more experienced workers in the same fields stayed essentially stable, a pattern several economists have pointed to as evidence this is not simply a broad hiring downturn but something concentrated specifically at the entry level.

The counter-evidence complicates the extinction narrative considerably though. IBM, under chief human resources officer Nickle LaMoreaux, tripled its junior developer intake in 2026, explicitly restructuring the roles so new hires spend less time on routine coding and more time on customer interaction and reviewing AI generated output, with LaMoreaux arguing bluntly that companies three to five years from now that succeed will be the ones that doubled down on entry-level hiring during this exact period rather than abandoning it. Cloudflare has continued hiring more than 1,100 early-career staff in 2026, and Shopify runs an internship program bringing in around 1,000 people annually. The split appears to run along fairly predictable lines, enterprise software vendors, financial institutions, healthcare platforms and other organizations with long-lived, complex internal systems continue hiring juniors because, as one analysis put it plainly, future senior engineers have to come from somewhere, while startups and pure software product companies have leaned much harder into hiring one senior engineer instead of two juniors. Anthropic has stated explicitly that it no longer needs junior engineers in the traditional sense, a genuinely notable position for a company at the center of building the tools driving this exact shift.

From writing code to reviewing and architecting it

Where this settles for the entry-level role itself seems to be less about a clean disappearance and more about a real redefinition of what junior even means. The Stack Overflow 2025 Developer Survey found 84 percent of developers now use AI coding tools, up 14 percentage points from 2023, and separate benchmarking found AI performance on SWE-Bench, a standard test of real world software engineering problems, improved from solving just 4.4 percent of problems in 2023 to 71.7 percent within roughly a year, a genuinely dramatic capability jump in a very short window. Companies report code output per sprint rising somewhere between 40 and 55 percent after adopting tools like GitHub Copilot, Cursor and Claude, and AI coding assistant company Cursor was recently valued at 60 billion dollars in a funding round, a figure that itself signals how much investor conviction has shifted toward tools that plan and self-correct code rather than tools that simply generate it faster.

The shift IBM's restructuring points toward, less time spent on routine implementation, more time spent reviewing AI generated output and interacting directly with the people the software is actually built for, mirrors what several analysts now describe as the genuinely new junior profile, someone who arrives already fluent in working alongside AI tools rather than being trained into that fluency on the job over their first two years. That is a real and uncomfortable inversion of how the career ladder used to work, where junior years were specifically the time you were expected to be inexperienced and slow while you learned, and increasingly the expectation is that new hires arrive pre-trained on skills the industry used to consider a manager's job to teach. A 2024 SHRM survey found 70 percent of hiring managers believe AI can now do the jobs interns used to do, and 57 percent said they trust AI's output more than that of an intern or recent graduate, numbers that help explain the hiring pattern even if they say something fairly bleak about how quickly institutional trust in inexperienced humans has eroded.

The mentorship gap nobody has solved yet

The genuinely unresolved risk sitting underneath all of this is what several researchers have started calling a missing generation problem. If a meaningful cohort of companies stops hiring and training juniors now, the pipeline that normally produces the senior engineers of five or ten years from now simply thins out, and nobody has a clean answer yet for where that experience comes from once the current senior cohort starts retiring or burning out. Forrester's 2026 predictions forecast a 20 percent drop in computer science enrollments as prospective students respond to the deteriorating entry-level job market signal, which risks becoming a self-reinforcing feedback loop, fewer students choose the degree because the entry-level job looks scarce, which shrinks the pipeline further just as the industry may eventually need it most. Whether the companies still investing seriously in junior hiring, IBM, Cloudflare, Shopify and others, prove to be the ones with the right long-term instinct, or whether AI capability keeps improving fast enough that the traditional junior role simply never comes back in its old form, is genuinely one of the more consequential open questions in the entire industry right now, and the honest answer is that nobody, including the companies making these bets, actually knows yet which way it resolves.

[Erin82]

This is a really thorough rundown of the data and I appreciate that it doesn't collapse into either the junior developer is dead or this is all overblown camp, but I think it slightly undersells how much the actual day to day work of a junior has already changed even at the companies still hiring. The IBM example is presented as a reassuring counterpoint, but read the actual restructuring closely, less routine coding, more reviewing AI output and customer interaction, that's not really the same job as junior developer used to describe, it's a genuinely different role that happens to still carry the same job title. I think the piece could have pushed harder on that point rather than treating IBM tripling its intake as straightforwardly good news for the traditional entry-level path. I'd also gently push back on how much weight the essay gives to the SWE-Bench jump from 4.4 to 71.7 percent, benchmark performance on curated test problems has a long history of not translating cleanly into messy real production codebases with legacy systems, inconsistent documentation and genuinely ambiguous requirements, so I'd want to see that number paired with something closer to real world deployment success rates before treating it as proof the technical capability gap is basically closed. Where I fully agree is on the missing generation risk, that's the part of this whole debate that gets the least attention relative to how serious it actually is, an industry quietly eating its own future senior talent pool to save money on juniors right now is a genuinely short-sighted trade if the pessimistic scenario plays out.

[Brittle Coder]

Strong piece overall, and I want to push back a bit on the previous reply's skepticism about IBM specifically, because I think reframing the junior role rather than eliminating it is actually the correct and honestly overdue response here rather than a disguised version of decline. The old junior developer job description, spend two years writing fairly mechanical boilerplate and small bug fixes while more senior people slowly teach you the bigger picture, was never some sacred format worth preserving for its own sake, it existed because that was the most efficient way to onboard someone given the tools available at the time. If AI can now handle a meaningful share of that mechanical work, restructuring the role around review, architecture and direct stakeholder interaction earlier in someone's career is arguably a better use of a junior's actual time and a faster path to real judgment than three years of writing CRUD endpoints ever was. Where I do think the essay slightly undersells the risk is the SHRM survey finding 57 percent of hiring managers now trust AI output over an intern's, because that number reveals something the piece doesn't fully unpack, a lot of that trust gap probably isn't really about AI being more competent than a smart 22 year old, it's about AI being more predictable and never having a bad week, and optimizing hiring decisions around predictability over raw potential is exactly the kind of short-term thinking that feeds directly into the missing generation problem the essay closes on. Both things can be true at once, the traditional junior role probably did need reinventing, and the specific way companies are currently doing that reinvention, driven mostly by immediate cost pressure rather than any careful long-term talent strategy, is genuinely risky for the industry a decade out.