The landscape of software development is undergoing a rapid transformation, evidenced by recent releases from OpenAI and Anthropic. Their new coding models, GPT-5.3-Codex and Claude Opus 4.6, have initiated a period of re-evaluation for many in the tech industry, prompting questions about the future of traditional programming roles. These advanced models are demonstrating capabilities that suggest a fundamental shift in how code is conceived, written, and deployed.
GPT-5.3-Codex, in particular, has showcased significantly improved performance on coding benchmarks, drawing attention for its ability to handle complex tasks. Not to be outdone, Claude Opus 4.6 introduces a novel feature: the deployment of autonomous AI agent teams. These teams are designed to tackle various aspects of large-scale projects concurrently, streamlining development processes. Both systems are proficient in writing, testing, and debugging code, even possessing the capacity to iterate on their own work and refine features before presenting finished results to developers. This self-sufficiency marks a considerable leap from previous AI tools.
The implications of these advancements were perhaps most sharply articulated by Matt Shumer, CEO of OthersideAI, whose viral essay captured the attention of many software engineers. Shumer contended that these models now manage the entire development cycle autonomously. He described AI systems writing extensive lines of code, launching applications, testing functionalities, and refining them until satisfactory, with human developers merely outlining desired outcomes. This perspective suggested that AI could disrupt employment more profoundly than even the COVID-19 pandemic. While Reddit co-founder Alexis Ohanian echoed similar sentiments, others, including NYU professor Gary Marcus, dismissed these claims as “weaponized hype,” citing a lack of empirical data to support the notion of AI flawlessly building complex applications. Fortune’s Jeremy Kahn further posited that coding’s inherent characteristics, such as automated testing, might make it uniquely susceptible to full automation compared to other knowledge-based fields.
Anecdotal evidence from within the industry suggests that for many engineers, Shumer’s observations are not entirely new. A growing number of developers report relying heavily on AI to generate code, effectively transitioning from direct coding to a more supervisory role. This trend has been slowly building over the past year as AI models became increasingly capable of handling intricate tasks independently. Developers at leading tech companies, while no longer writing every line of code themselves, continue to build software. Their expertise has evolved to architecting solutions and guiding AI systems, rather than the minute details of syntax. The latest models, some argue, simply brought this ongoing transformation to broader public awareness.
Spotify’s co-CEO, Gustav Söderström, provided a compelling example of this paradigm shift during a recent earnings call. He noted that the company’s top developers had not written a single line of code since December. Spotify’s internal system leverages Claude Code for remote deployment, allowing engineers to direct AI to address bugs or integrate new features via Slack, even while commuting. This workflow enables them to merge completed work into production before reaching the office. Söderström stated that Spotify shipped over 50 new features in 2025 using these AI-driven methods.
Even within Anthropic itself, engineers are extensively utilizing their own tools for code generation. Boris Cherny, head of Claude Code, recently indicated he had not personally written code in over two months. Anthropic previously informed Fortune that 70% to 90% of the company’s code is now AI-generated. A notable milestone has also been reached: these models are now playing a material role in their own evolution. OpenAI revealed that GPT-5.3-Codex was “instrumental in creating itself,” signifying a new phase in AI development. Similarly, Cherny’s team at Anthropic developed Claude Cowork, a non-technical file management version of Claude Code, in roughly a week and a half, largely using Claude Code itself. Approximately 90% of Claude Code’s own codebase, according to Cherny, is now written by Claude Code.
Despite the clear gains in productivity, some developers are beginning to caution about potential downsides, including burnout. Veteran engineer Steve Yegge highlighted concerns that these powerful AI tools could inadvertently lead to overwork. In a widely circulated blogpost, Yegge described instances of falling asleep unexpectedly after extended coding sessions and colleagues considering the installation of nap pods. He argued that the highly efficient nature of AI coding tools could push developers toward unsustainable workloads. While acknowledging the “10x boost” in productivity these tools offer, Yegge emphasized that “building things with AI takes a lot of human energy,” suggesting a new form of cognitive demand on developers.
