Coding with LLMs in the summer of 2025 – an update (antirez.com)

Article summary

The article discusses the author's experience with using large language models (LLMs) for coding, specifically Gemini 2.5 PRO and Claude Opus 4. The author highlights the benefits of using LLMs, such as eliminating bugs, exploring ideas, and accelerating work, but also notes the importance of human involvement and control. The author emphasizes the need for clear communication, providing large context, and using the right LLMs to achieve maximum quality of work.

Main themes

• LLMs for coding
• Human involvement and control
• Context and prompt engineering
• LLM choice and configuration
• Automated code review
• PhD-level knowledge and research

What commenters say

• Using LLMs for coding can be beneficial, but it requires human involvement and control to produce high-quality work.
• The effectiveness of LLMs depends on the quality of the context provided, with both insufficient and excessive context being detrimental to performance.
• Some commenters disagree on the choice of LLM, with some preferring Sonnet over Opus due to its lower cost and better performance in certain tasks.
• The use of LLMs for automated code review can be effective, but it requires careful prompt engineering and consideration of the team's tribal knowledge and standards.
• The concept of 'PhD-level knowledge' is misleading, as the primary purpose of a PhD is to learn how to conduct research, not simply acquire knowledge.
• LLMs are not yet capable of exploring hypotheses or asking questions on their own, and require human guidance to produce meaningful results.
• The choice of LLM and its configuration can significantly impact the quality of the output, with some models being more prone to producing low-quality code or going off the rails.