Is traditional coding dead? That’s the question many developers have been asking themselves this week following the launch of powerful new coding models from OpenAI and Anthropic.
Last week, OpenAI and Anthropic dropped their respective coding models—GPT-5.3-Codex and Claude Opus 4.6—both of which represented significant leaps in AI coding capabilities. GPT-5.3-Codex showed markedly higher performance on coding benchmarks than earlier models, while Opus 4.6 introduced a feature that lets users deploy autonomous AI agent teams that can tackle different aspects of complex projects simultaneously. Both models can write, test, and debug code with minimal human intervention—even iterating on their own work and refining features before presenting results to developers.
For many engineers, some of Shumer’s warnings just reflect their current reality. Many engineers say they have stopped coding entirely, instead relying on AI to write code at their direction.
While the new releases do represent meaningful improvements, developers also said the industry has been undergoing a slow transformation over the past year as models became capable enough to handle increasingly complex tasks autonomously. While developers at leading tech companies have largely stopped writing code line-by-line, they haven’t stopped building software—they’ve become directors of AI systems that do the typing for them. The skill has transformed from writing code to architecting solutions and guiding AI tools. The new models, some argue, mainly “burst the bubble” around AI coding by making people outside coding aware of a trend engineers have been experiencing for months.
The models themselves have also reached a recursive milestone: They’re now materially helping to build more advanced iterations of themselves. OpenAI said GPT-5.3-Codex “is our first model that was instrumental in creating itself,” a significant shift in how AI development works. Similarly, Anthropic’s Cherny said his team built Claude Cowork—a non-technical version of Claude Code for file management—in approximately a week and a half, largely using Claude Code itself. Even for Claude Code, Cherny said about 90% of its own code is now written by Claude Code.
In a widely shared blogpost, Yegge described falling asleep suddenly after long coding sessions and colleagues considering installing nap pods at their office. The addictive nature of AI coding tools, he argues, is pushing developers to take on unsustainable workloads. “With a 10x boost, if you give an engineer Claude Code, then once they’re fluent, their work stream will produce nine additional engineers’ worth of value,” he wrote. But “building things with AI takes a lot of human energy.”
