The debate over whether AI will replace developers has officially moved from theoretical to practical. Boris Cherny, the creator of Claude Code, revealed that for the first time in his career, he did not open an IDE last month. Instead, Opus 4.5 generated approximately 200 pull requests, with every single line of code written by the AI. This milestone, reported by The Pragmatic Engineer, signals that the industry has crossed a critical threshold where AI is no longer just an autocomplete tool, but a primary code generator for even the most advanced engineering teams.

The Late 2025 Model Leap

This shift wasn't gradual; it was a sudden step change driven by three major model releases in November and December. Google launched Gemini 3 on November 17, followed by Anthropic’s Opus 4.5 on November 24, and OpenAI’s GPT-5.2 on December 11. These models didn't just improve incrementally; they tipped over an invisible capability line. Simon Willison, a respected independent expert, noted that GPT-5.2 and Opus 4.5 suddenly opened up much harder coding problems that were previously unsolvable by agents. Peter Steinberger, creator of PSPDFKit, confirmed that the jump from GPT-5.1 to 5.2 was 'massive,' reducing the need for custom CLI agents to get unstuck from multiple times a day to just a few times a week.

Skeptics Convert to Believers

Even prominent skeptics have revised their stance. Andrej Karpathy, a co-founder of OpenAI who previously called AI coding tools 'slop' in October, posted on December 26 that he had 'never felt this much behind as a programmer.' He described the new abstraction layer as a 'powerful alien tool' that everyone must learn to operate while the profession undergoes a magnitude 9 earthquake. Similarly, DHH, the creator of Ruby on Rails, admitted that his resistance to AI coding was based on earlier models being insufficient. He now argues that spending time rewriting AI output is faster than writing code from scratch, marking a complete flip in his perspective since mid-2025.

The Death of Syntax and Specialization

As AI handles the implementation details, the value of traditional engineering skills is being redefined. Malte Ubl, CTO at Vercel, stated that 'the cost of software production is trending towards zero,' noting he built two major open-source projects and started a book during the holiday break using Opus 4.5. The article argues that being a language polyglot is becoming less valuable because AI allows engineers to jump into any codebase and implement features without deep prior expertise. Roles requiring specific language knowledge, such as separate frontend and backend developers, are likely to merge, especially in startups where a single specialist can leverage AI to unblock themselves across the entire stack.

Key Takeaways

  • Boris Cherny committed 200 PRs written entirely by Opus 4.5 without opening an IDE.
  • Major model releases (Opus 4.5, GPT-5.2, Gemini 3) created a sudden capability inflection point in late 2025.
  • Andrej Karpathy and DHH have shifted from skepticism to acknowledging AI's dominance in code generation.
  • The value of syntax memorization and narrow language specialization is declining rapidly.
  • Product managers and designers are increasingly capable of building functional prototypes without engineer support.

The Bottom Line

If the creator of Claude Code isn't writing code by hand, your IDE habits are already obsolete. The bottleneck is no longer typing speed or syntax recall, but your ability to architect systems and define precise prompts.