Vibe Coding: AI’s Next Act Puts Software Engineers on Edge
Artificial intelligence has been writing snippets of code for more than two years, but 2025 is shaping up as the moment it graduates from sidekick to co-author. Rapid advances in “vibe coding” — the practice of building entire apps by prompting large language models (LLMs) with plain English — are fuelling predictions of a seismic shift in software work and stoking anxiety across developer forums.
From Autocomplete to App Builder
When ChatGPT arrived in late 2022, coding aids like GitHub Copilot could finish a function or suggest a line. Newly released agentic models from OpenAI, Anthropic and Google can now generate project plans, spin up test suites and iterate thousands of lines in minutes. Steve Yegge, a former Google engineer now at code-search firm Sourcegraph, says he routinely delegates four separate projects to AI while he “just burns tokens.”
The pivot has given rise to start-ups such as Cursor and Windsurf (rumoured to be acquisition bait for OpenAI) and popularised the term “vibe coding,” coined by AI researcher Andrej Karpathy to describe the looser, prompt-driven style of software creation.
A Coming Job Crunch — or an Uber Moment?
Anthropic CEO Dario Amodei told a Council on Foreign Relations audience in March that AI could be writing “90 percent of the code” within six months and “essentially all” of it a year later. That has some engineers forecasting a jobs apocalypse.
MIT labour economist David Autor isn’t convinced the profession will vanish but warns the impact could mirror ride-hailing’s disruption of taxis: more code produced at lower prices and downward pressure on mid-level wages. Firms like Databricks already say they hire fewer “average” developers; others stress they still need experts for architecture, security and oversight.
Hype Meets Hard Reality
Critics counter that today’s models remain flaky. AI-generated code can hallucinate, expose security holes or rack up costly cloud bills. Christine Yen, CEO of observability platform Honeycomb, says her team sees at best a 50 percent productivity bump on routine tasks, while anything mission-critical still demands human judgment. “AI just isn’t good enough yet to be additive,” she says.
Veteran coders like Ken Thompson at Anaconda argue that vibe coding resembles past leaps — from assembly to high-level languages — in abstracting complexity. But unlike deterministic compilers, LLMs offer no guarantee they will repeat the same answer twice.
The Skills Premium Shifts
Even AI optimists acknowledge that guardrails are thin. Yegge calls LLMs “toddlers you must watch constantly,” and he and co-author Gene Kim urge developers to embrace modular designs, constant testing and strict cost monitoring. The real casualty, they say, may be complacency: those who fail to adopt AI will simply fall behind.
For now, the safest career path may be learning how to steer the models rather than competing with them. As Milestone CEO Liad Elidan puts it, “The demand for code is still growing. The bar for what counts as a great engineer is just rising faster than before.”
Whether vibe coding leads to mass layoffs or a broader, cheaper software renaissance, one thing looks certain: the profession that once epitomised technical security is entering its most volatile chapter yet.
Photo Credit: DepositPhotos.com