In 2025, the author underwent a significant shift in perspective on large language models (LLMs). Initially skeptical, viewing them as clumsy tools unsuitable for real software development, they now rely on Claude Code for about 90% of their coding work. The change was prompted by improvements in LLM capabilities, enabling more efficient workflows where the author could specify tasks and let the AI handle the grunt work. Despite lingering issues with bugs and inaccuracies, the author has found that tools like Cursor Bugbot can catch errors that they might overlook, prompting them to rethink their role as a programmer. The discourse surrounding LLMs has become increasingly polarized, with strong opinions on both sides—either heralding AI as a game-changer or fearing its potential to disrupt the industry. The author notes that many discussions have devolved into tribalism, clouding judgment. They reflect on the initial hype around LLMs, noting that those they once dismissed as lacking credibility are now at the forefront of advocating for AI in programming. The author acknowledges the various concerns raised by peers regarding security, performance, and accessibility but counters that these models can address some of these issues through iterative improvements. While grappling with the evolving landscape of software development, the author emphasizes that no one truly knows where things are headed. They advocate for experimentation and curiosity rather than despair or denial. The landscape has changed dramatically in just three years, and the author suggests that developers remain open-minded and empathetic to each other’s experiences, regardless of their stance on the integration of AI into their work.