How thorough is Claude with planning a small app, really? My second opinion with Gemini revealed some holes.
retained text
Hello. I'm currently planning a relatively minimal web based calculator for an industry I know nothing about. I've spent a good deal of time with Claude just researching the industry and how workers calculate their earnings. It's revealed many nuances that need special attention when calculating. However, in my chat with Claude it always felt like it wasn't able to thoroughly source the information I asked for and it left many questions on the table. I decided to check with Gemini. I'm only using the free version that's part of the search option in browser. Not the actual Gemini app. Upon giving Gemini my latest configuration, just explaining it, no code, and the questions I'm left with, it was able to discover a lot more than Claude did. Gemini found sources that answered my questions and filled the gaps, it was also more thorough with its own assessment of the calculator concept and brought up new concerns or even some flat out broken conditions, and it actually provided sound fixes for each. For example, there's an earnings field, but the user might enter a value that is either before or after a commission taken. And if the value is marked as after, Gemini noticed that the Gross total output does not account for that commission percentage. I didn't get anything like this from Claude. I've used Claude and Claude code for a couple years on a handful of projects with decent success. But this level of thorough output has been a problem with Claude and typically takes me several prompts and attempts to get things ironed out. I have to do a lot of the thinking to recognize possible issues, rather than Claude noticing things on the first try. I'm using sonnet 5 on high with extended thinking. To be fair, Claude is "thorough" with its length of output and I haven't had many problems like this on other projects. So maybe it's this specific calculator concept that's causing trouble. I'm wondering if others have had this experience. Should I always use another LLM to double check things or get second opinions? Thanks!
document fetched 2026-09-12 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1wdm8nh/how_thorough_is_claude_with_planning_a_small_app