🎧 Listen to this article
Narrated by Talon · The Noble House
In February 2025, Andrej Karpathy — co-founder of OpenAI and former head of AI at Tesla — posted a description of a new practice he was calling "vibe coding": you describe what you want in plain language, let the AI write the code, and accept changes based on whether things work rather than reviewing every line. "You fully give in to the vibes," he wrote, "embrace exponentials, and forget that the code even exists."
Collins Dictionary named it their Word of the Year for 2025. Merriam-Webster listed it in March 2025 as a "slang and trending" expression. MIT Technology Review listed generative coding as a 2026 breakthrough technology. By Observer's February 2026 coverage, Karpathy had already moved on to a new term — vibe coding had become mainstream enough that the person who coined it was already describing the next thing.
The Printing Press Parallel Is More Precise Than It Sounds
Software creation has been a monastic profession for sixty years. If you wanted software, you either hired someone who could write code, or you learned to write it yourself. The barrier was real and high: syntax, data structures, debugging, architectural patterns. Most people had legitimate reasons to create custom software and no accessible path to do so.
The printing press parallel isn't about cheaper books. It's about what happens to literacy when the cost of using the technology drops to near zero. Medieval monks were literate not just because they needed to write — they were literate because the skill was rare enough to create professional value. When Gutenberg's press made books abundant, literacy stopped being a monastic privilege and became a general skill. The economic incentive to learn changed; so did the population of people who learned.
Vibe coding decouples the ability to create software from the ability to write code. That's the inversion: for sixty years, "can you build software" and "can you write code" were functionally identical questions. They're no longer the same question.
The Evidence Isn't Anecdotal
Replit launched Agent 3 in late 2025, positioned explicitly for non-technical users. Their reported data showed non-developer usage growing faster than developer usage quarter-over-quarter. Y Combinator's 2025 cohort included companies with zero full-time engineers building products that passed user acquisition milestones — the first time that had happened in the accelerator's history, per multiple coverage reports.
The use cases emerging from non-developer builders are specific and practical: marketing analytics tools built by marketing professionals, real estate deal analyzers built by investors, compliance tracking systems built by lawyers who wanted something their IT department couldn't prioritize. These aren't toys — they're functional software built by domain experts who previously had no path to software creation.
The quality variance is real and documented. Non-developer vibe coding frequently produces software with security vulnerabilities, edge case failures, and architectural choices that professional engineers would avoid. The software works in the demo and fails at scale. This is not a permanent state — it's the 2026 starting point of a capability curve that will improve — but it's important to acknowledge the gap rather than paper over it.
What Professional Software Engineers Actually Face
The professional software engineering response to vibe coding splits into two camps that are both partially right.
Camp one: "This threatens junior roles but not senior ones." The argument is that senior engineers have judgment — they understand why certain architectural decisions create debt, where AI-generated code is likely to fail, how to structure systems that survive contact with real users. AI tools amplify that judgment rather than replace it. This camp is correct about senior engineers and incorrect that the junior role disruption doesn't matter (see: the apprenticeship pipeline problem).
Camp two: "Non-developer builders will discover the limits quickly and return to professionals for anything serious." This has some validity — complex software has irreducible complexity that requires technical expertise to navigate. But "serious" is a lower bar than camp two assumes. A 200-user internal analytics tool built by a marketing manager and maintained by that manager is "serious" in the sense of providing real business value, without ever requiring professional engineering involvement.
The resolution: vibe coding expands the total volume of software creation. Most of that expansion happens in the "good enough for our actual use case" segment that professionals never would have served anyway. The net effect is more software in the world, built by more people, at different quality tiers. This isn't a threat to professional software engineering; it's a structural change in what the profession serves.
The Skill That's Actually Changing
If vibe coding makes software creation accessible to non-developers, the constrained resource shifts from "can you write code?" to "do you understand the problem well enough to specify the solution?" Domain expertise becomes more valuable relative to syntax knowledge. The person who deeply understands the business problem they're solving has a larger portion of the needed capability than they did when they also needed to know Python.
This is a genuine inversion from the previous sixty years. It doesn't eliminate coding as a valuable skill — it repositions coding as one input into software creation rather than the primary input. The professionals who thrive in this environment are those who understand both the domain they're building for and the technical constraints that determine whether a solution will hold up. Neither alone is sufficient. Both together are increasingly powerful.
Sources: Andrej Karpathy on X, February 2025 (original "vibe coding" post); Wikipedia, "Vibe Coding" article; Collins English Dictionary Word of the Year 2025 announcement; Merriam-Webster "slang and trending" listing, March 2025; Observer, "Vibe Coding Inventor Andrej Karpathy Has a New Term," February 2026; MIT Technology Review, generative coding as 2026 breakthrough technology; Replit Agent 3 launch and non-developer usage data, late 2025
