Cursor vs Droid: Only One Did the Audit
Several hundred dollars later, I discovered something most developers don't realize: the same AI model with the same prompt behaves completely differently depending on which IDE you use. I ran GPT-5.1 Codex Max and Opus 4.5 through both Cursor and Droid using identical prompts on a real backend migration. The finding? In Cursor, all the Codex Max models skipped the file audit and jumped straight to planning. In Droid, that same model actually completed the audit first. Same prompt. Same model. Different IDE instructions changed everything. This video also shows a technique I call "prompting without being a prompt engineer" - using Opus 4.5 to rewrite your unstructured thoughts into properly formatted prompts before feeding them to other models. The structured output is significantly better than what you'd write manually. Join my Discord: https://rfer.me/discord 1-1 AI Shipping Mentorship (December): https://startmy.ai My LLM Rules and Prompts Repo: https://github.com/rayfernando1337/llm-cursor-rules CONNECT WITH RAY X (Twitter): https://x.com/RayFernando1337 Key moments 00:00 Comparing AI models across different IDEs for a real-world backend rewrite 00:59 Using one AI (Opus) to create a structured prompt from unstructured thoughts 02:32 Demonstrating the technique of using an AI to rewrite your own prompts 03:56 Testing the structured prompt in the Droid terminal using spec mode 05:37 Analyzing the detailed multi-step migration plan generated by Droid 06:34 Comparing how Cursor and Droid handle the same task with the same model 08:54 Key finding: The IDE's instructions significantly alter the AI model's behavior 10:20 The philosophy behind using AI for advanced planning in software engineering 11:03 How to iterate on AI-generated plans by reapplying learned knowledge 12:03 Final thoughts and a call for community feedback on their planning workflows #Cursor #Droid #FactoryAI #GPT5 #Opus45 #AIcoding #SoftwareEngineering #RayFernando #PlanMode #SpecMode
