Tips from Salvatore Sanfilippo aka antirez (creator of redis, kilo, etc.):

Advanced language models like Gemini 2.5 PRO and Claude Opus 4 can significantly boost programmer productivity — if used correctly.

What LLMs can do for you

• Early bug detection. They catch bugs before users see them.
• Rapid prototyping. Let them write quick temporary code to test ideas.
• Paired design. You bring experience; the model brings encyclopedic knowledge.
• Writing well-defined parts. Give them clear tasks, get working code.
• Filling skill gaps. They help you pick up tools or languages you barely know.

How to work with them

1. Stay in control. For anything beyond a small script, combine your judgment with the model’s output. Code generated only by an LLM often ends up bloated and brittle.
2. Provide full context. Paste relevant code, documentation, and your notes:

• Bad solutions to avoid — and why.
• Promising ideas to explore.
• Exact goals, constraints, and style rules. More context = better answers.

3. Use the best models. Stick with Gemini 2.5 PRO and Claude Opus 4. Work with the base model, not diluted “agents” or editor plugins that hide context.
4. Keep the loop short. Manually move code between editor and chat. Check every change yourself.

Conclusion Right now, human + LLM is better than either alone. Use models to speed up and improve your work, but keep control and ownership of the code. Periodically test fully automated agents; switch only when they clearly outperform you. Until then, avoid the hype, use AI wisely, and don’t fall behind by ignoring it.