𝗦𝘁𝗼𝗽 𝗨𝘀𝗶𝗻𝗴 𝗔𝗜 𝗔𝘀 𝗔 𝗦𝗲𝗮𝗿𝗰𝗵 𝗘𝗻𝗴𝗶𝗻𝗲
I watched a developer spend forty minutes talking to Claude about a database schema. Every message was a question. Every reply was an answer. At the end, they had a schema that worked but felt wrong.
The model was not the problem. The format was. They used a conversation like a Q&A session. They needed a thinking session. These are not the same.
Google trained you to use queries. You form a short term, get results, and close the tab. This habit fails with AI. When you treat an AI like a search engine, you only get answers. You do not get insights. You get syntax instead of architecture. You get what you asked for instead of what you need.
Search engines find existing facts. AI can reason through new problems. A search engine is a library. An AI can be a colleague. A library gives you what is written. A colleague pushes back on your assumptions and finds holes in your plan.
Look at these two ways to ask the same thing:
Search mode: "What is the best way to structure a Python microservice?"
Thinking mode: "I am building a Python microservice for three APIs. Each API has different retry rules and data shapes. I am choosing between one FastAPI app with a queue or three separate consumers. Two engineers will maintain this. What tradeoffs am I missing?"
The first way gets you a blog post. The second way makes you think.
The best users do not have clever prompts. They bring more context. They describe what they build, their constraints, and what they already tried.
They also say things like: "This feels wrong but I cannot say why."
This is a great prompt. It tells the model to probe your assumptions. It helps the model find the problem you cannot name.
Shallow results happen when questions are too clean. Real engineering is messy. It has deadlines and legacy code. If you strip the mess away, the AI gives you a clean answer that ignores reality. The mess is the actual problem.
Before your next prompt, spend two minutes writing these three things:
- What you want to accomplish.
- The approach you are considering.
- What you are uncertain about.
Give all three to the model first.
This takes more effort. It is the effort you should do before you write any code. The AI does not add this step. It just makes skipping it more expensive.
Un moteur de recherche a besoin d'une requête propre. Un partenaire de réflexion a besoin d'une vue d'ensemble.
Cessez de nettoyer le désordre. Le désordre est votre contexte.
Source : https://dev.to/gmoustakas/stop-using-ai-as-a-search-engine-with-extra-steps-49oo
Communauté d'apprentissage optionnelle : https://t.me/GyaanSetuAi