Recent advancements in large language models (LLMs) have shifted the coding landscape dramatically. The author notes a transition from primarily manual coding with some autocomplete assistance to relying heavily on LLMs for coding tasks. In just a month, their workflow changed to 80% agent coding, with only 20% dedicated to manual edits. This shift highlights the growing trend among engineers to program in English, instructing the LLMs on what to write. While this has streamlined the coding process, it also raises concerns about reliance on agents, as they can produce subtle errors and make incorrect assumptions without proper oversight. The author emphasizes the need for caution when using LLMs, noting that while they can significantly enhance productivity, they still make mistakes. These errors often stem from complex conceptual misunderstandings rather than simple syntax issues. The models tend to overcomplicate solutions and may even alter or remove comments and code that they find unappealing. Despite these challenges, the author finds it hard to imagine returning to a purely manual coding approach. They describe a new dynamic where working with LLMs feels more enjoyable, as it eliminates some of the repetitive tasks and allows for more creative problem-solving. The article also reflects on the potential implications for the future of coding. The author speculates about the rise of the "10X engineer" and whether generalists might outperform specialists with LLM assistance. They predict a significant transformation in software engineering by 2026, driven by the advanced capabilities of LLMs like Claude and Codex. The author expresses concerns about a potential "slopacolypse" in digital media, where the quality of content generated by LLMs may decline. Ultimately, the text captures a moment of transition and adaptation in the tech industry, as professionals navigate the benefits and challenges posed by these powerful tools.