The artificial intelligence landscape is currently marked by a significant perception gap, a phenomenon that is disproportionately affecting the software development community. While the layperson might encounter AI through consumer-facing applications that exhibit limitations, developers, particularly those working with advanced models, are experiencing a profound and rapidly accelerating transformation in their professional capabilities. This disparity is not merely a technical nuance; it signals the initial wave of a broader industrial revolution driven by AI, with developers serving as the early adopters and harbingers of change across numerous sectors.
The core of this divergence, as articulated by OpenAI co-founder Andrej Karpathy in a recent essay, lies in the differing levels of engagement with AI technology. Karpathy posits that individuals who have only interacted with free, foundational AI models, such as earlier versions of ChatGPT, may have formed opinions based on observed "hallucinations" and suboptimal outputs. This segment of the population, therefore, perceives AI as a tool with significant limitations. In contrast, developers who leverage paid access to frontier models like OpenAI’s Codex or Anthropic’s Claude Code, and integrate them into their professional workflows, are witnessing "nothing short of staggering" advancements. These professionals are observing AI systems capable of solving complex programming challenges that previously required days or weeks of human effort, leading to a sentiment Karpathy terms "AI Psychosis" due to the sheer, almost unbelievable, pace of improvement.
While Karpathy’s technical explanation, focusing on the efficacy of reinforcement learning with verifiable reward functions in domains like coding and mathematics, is accurate, it overlooks a crucial element: the confluence of AI capability, AI expertise, and deep domain knowledge within the developer community itself. Developers are not just users of these advanced tools; they are often the architects and implementers of the systems that integrate them. This intimate relationship allows them to push the boundaries of AI application in ways that are not yet replicable in other professional fields. As The New Stack has been tracking since Karpathy’s early discussions on "vibe coding" evolving into "agentic engineering," the focus has shifted from a consumer-centric view to one that emphasizes the developer’s role in orchestrating these increasingly sophisticated AI agents. This inherent overlap is why developers are experiencing the full impact of AI transformation first.
The widening chasm in perception is a direct consequence of this specialized integration. While a journalist or editor might explore AI’s potential in content creation, the tools are not yet as finely tuned for their specific domain as they are for code. The current optimization for software development stems from the fact that AI models, particularly those adept at reasoning and code generation, benefit from the structured, logical, and often mathematically definable nature of programming tasks. The "verifiable reward functions" Karpathy mentions are abundant in software development, allowing for more effective training and refinement of AI models. This has led to a concentrated effort by AI developers to create tools that directly address the pain points and enhance the productivity of programmers.
Why Everyone Else Will Feel It Next
The trajectory of AI adoption suggests that the transformative effects currently concentrated within the developer sphere are poised to expand rapidly across other industries. A prime example of this expansion is Anthropic’s recent move of Claude Cowork from a research preview to general availability, complete with enterprise-grade features. This includes plugins, managed agents, and connectors for widely used business applications like Google Drive, Gmail, and DocuSign, alongside industry-specific tools. The underlying philosophy of Claude Cowork is a direct extension of Claude Code’s success: to revolutionize professional workflows beyond software development. This indicates a deliberate effort to equip professionals in fields such as Human Resources, Operations, Finance, and Design with agentic tools that mirror the capabilities currently astonishing developers. While Claude Code saw a significant period of development before its widespread impact became apparent in late 2025, Claude Cowork could follow a similar, albeit potentially accelerated, adoption curve.
This impending shift is amplified by emerging societal attitudes towards AI. A recent Gallup study highlighted a notable increase in negative sentiment among Generation Z regarding AI. The study found that 31% of Gen Z respondents now express anger towards AI, a nine-percentage-point increase from the previous year, while excitement has waned significantly. Furthermore, 80% of this demographic believe AI may hinder their future learning capabilities. The New York Times characterized this as a generational mood shift, linking it to pervasive exposure to low-quality AI outputs and institutional confusion. Teenagers entering college are reportedly resistant to fully embracing AI, often reacting to experiences with chatbots that "hallucinate" or the proliferation of AI-generated content, alongside conflicting institutional policies. This contrasts sharply with the sophisticated, domain-specific AI applications that are actively restructuring codebases and identifying critical security vulnerabilities for developers.
The economic implications of AI are becoming increasingly stark, with potential job displacement emerging as a significant concern, particularly for younger workers. Goldman Sachs estimates that AI is currently eliminating approximately 16,000 net U.S. jobs per month. This figure is a net result of roughly 25,000 jobs automated by AI and 9,000 jobs created through AI augmentation. Younger members of Generation Z are disproportionately represented in roles susceptible to automation, such as data entry, customer service, legal support, and billing. The impact is not confined to entry-level positions; a report by Nikkei Asia indicated that in the first quarter of 2026, nearly 80,000 tech workers were laid off, with a substantial portion attributed directly to AI. Cognizant’s chief AI officer suggested that the full impact on the workforce may take up to a year to materialize, emphasizing the nascent stage of this technological disruption. The underlying message is clear: while direct job loss to AI might be rare, individuals who fail to adapt and enhance their skills with AI tools are at risk of being outcompeted by those who do. This reiterates the core thesis of adapting to and leveraging AI to remain competitive in the evolving job market.
The Evidence of Spreading Impact
The diffusion of AI’s transformative power beyond its initial domain is becoming increasingly evident, particularly when advanced AI capability intersects with deep industry expertise. Anthropic’s recent unveiling of Claude Mythos Preview offers a compelling illustration of this phenomenon. Described by Anthropic as "far ahead of any other AI model in cyber capabilities," Mythos demonstrated an unprecedented ability to identify and exploit zero-day vulnerabilities across major operating systems and web browsers during internal red team assessments. The model discovered a 27-year-old denial-of-service bug in OpenBSD and a 16-year-old flaw in FFmpeg. Furthermore, it generated 181 functional exploits for Firefox JavaScript engine vulnerabilities, a stark contrast to the 2 exploits identified by the previous Opus 4.6 model.
A critical observation from Anthropic’s announcement is that these advanced cybersecurity capabilities were not explicitly trained into Mythos. Instead, they emerged as a "downstream consequence of general improvements in code, reasoning, and autonomy." This highlights a fundamental mechanism of AI diffusion: by enhancing a model’s general reasoning and coding abilities, specialized and highly valuable capabilities in adjacent fields, such as cybersecurity, law, medicine, and finance, can cascade organically. This emergent property suggests that as AI models become more broadly capable, their impact will extend far beyond their initial design parameters.
Anthropic has opted not to release Mythos publicly, instead launching Project Glasswing in collaboration with major technology partners including Amazon, Apple, and Microsoft. This initiative pledges $100 million in credits and $4 million in direct donations to bolster open-source security efforts, focusing on the defensive application of the model’s capabilities. The New York Times aptly described this development as a "cybersecurity reckoning," underscoring the profound implications for digital security.
Meanwhile, Meta’s internal AI adoption offers a glimpse into the aggressive integration of AI within large enterprises. An employee developed an internal dashboard, "Claudeonomics," to track AI token consumption across the company’s workforce of over 85,000 employees. The data revealed a staggering 60 trillion tokens consumed in just 30 days, with top users achieving astronomical consumption rates. Meta’s CTO, Andrew Bosworth, reportedly cited a top engineer achieving tenfold productivity gains by dedicating their entire salary to AI token usage. While concrete metrics linking token consumption directly to business outcomes remain elusive, the existence of a competitive leaderboard for AI usage among tens of thousands of employees signals a strong organizational belief in the productivity gains offered by advanced AI models. As these tools mature and become more accessible across various industries, the "token legends" of today, predominantly engineers, will likely be joined by professionals from a diverse range of fields.
The narrative is clear: developers are not an isolated group facing unique threats from AI. Instead, they represent the vanguard, experiencing the initial wave of a technological revolution that is fundamentally reshaping industries. Their early adoption and deep integration of AI tools position them at the forefront of this transformation, and as these capabilities mature and spread, other professions will inevitably follow suit. The challenge for individuals and organizations alike lies not in resisting this shift, but in understanding its trajectory and actively participating in the AI-powered future.