How to keep enjoying programming in a world of LLMs
4 hours ago
- The author advocates for writing most of your own code to maintain ownership, skill, and joy, using LLMs for planning, research, and cleanup instead of full code generation.
- LLMs should be used as bookkeeping and research tools, but humans should make crucial decisions and understand the domain at least as well as the AI.
- The author recommends a workflow where the human does the coding while the LLM handles planning, background research, and identifying potential pitfalls, keeping the developer in control.
- Use automated review cycles for all LLM-generated artifacts (code and plans) to catch errors and logical holes, and consider having your own code reviewed by an LLM as well.
- Avoid relying on frontier models due to high token costs, environmental impact, and lack of accountability; prefer smaller, open-weight models when possible.
- Treat token depletion as a service outage rather than your fault, and always have offline work (like a planned todo list) ready to continue without LLM access.
- Be mindful of LLM gibberish affecting mental health, take breaks, and prioritize human-to-human communication over AI-generated PR descriptions or messages.