[{"data":1,"prerenderedAt":119},["ShallowReactive",2],{"navigation":3,"\u002Fblog\u002Fbanksy-what-not-to-do-with-ai":25,"\u002Fblog\u002Fbanksy-what-not-to-do-with-ai-surround":116},[4],{"title":5,"path":6,"stem":7,"children":8,"page":24},"Blog","\u002Fblog","blog",[9,13,17,21],{"title":10,"path":11,"stem":12},"[object Object]","\u002Fblog\u002Fbanksy-what-not-to-do-with-ai","blog\u002Fbanksy-what-not-to-do-with-ai",{"title":14,"path":15,"stem":16},"Making Sense During the AI Revolution","\u002Fblog\u002Fmaking-sense-during-ai-revolution","blog\u002Fmaking-sense-during-ai-revolution",{"title":18,"path":19,"stem":20},"RAG vs AI Wiki","\u002Fblog\u002Frag-vs-ai-wiki-evolution-of-context","blog\u002Frag-vs-ai-wiki-evolution-of-context",{"title":10,"path":22,"stem":23},"\u002Fblog\u002Fvector-storage-in-rag-choosing-right-backend","blog\u002Fvector-storage-in-rag-choosing-right-backend",false,{"id":26,"title":10,"author":27,"body":28,"date":27,"description":34,"extension":101,"image":27,"meta":102,"minRead":27,"navigation":103,"path":11,"seo":104,"stem":12,"__hash__":115},"blog\u002Fblog\u002Fbanksy-what-not-to-do-with-ai.md",null,{"type":29,"value":30,"toc":93},"minimark",[31,35,38,43,56,60,71,75,86,90],[32,33,34],"p",{},"I refactored Alex into my personal banking assistant under strict time constraints.",[32,36,37],{},"It worked enough for a demo, but the approach created avoidable complexity.",[39,40,42],"h2",{"id":41},"what-i-did","What I did",[44,45,46,50,53],"ul",{},[47,48,49],"li",{},"Reused an existing codebase optimized for a different architecture.",[47,51,52],{},"Pushed rapid changes with large context windows and mixed-model workflows.",[47,54,55],{},"Prioritized delivery over structural cleanup.",[39,57,59],{"id":58},"problems-i-hit","Problems I hit",[44,61,62,65,68],{},[47,63,64],{},"Legacy resource names leaked into new domains.",[47,66,67],{},"Old workflow fragments remained active after redesign.",[47,69,70],{},"Infrastructure refactors required deeper recoding than expected.",[39,72,74],{"id":73},"what-i-would-do-differently","What I would do differently",[44,76,77,80,83],{},[47,78,79],{},"Start with an architecture map and deletion plan.",[47,81,82],{},"Rebuild core Python flows from clean boundaries.",[47,84,85],{},"Re-validate Terraform modules against the target domain before reuse.",[39,87,89],{"id":88},"takeaway","Takeaway",[32,91,92],{},"Plan first, then accelerate. AI tools amplify your approach, including flawed assumptions.",{"title":94,"searchDepth":95,"depth":95,"links":96},"",2,[97,98,99,100],{"id":41,"depth":95,"text":42},{"id":58,"depth":95,"text":59},{"id":73,"depth":95,"text":74},{"id":88,"depth":95,"text":89},"md",{},true,{"title":105,"description":34},{"Banksy":106,"description":107,"date":108,"image":109,"minRead":110,"author":111},"What Not to Do with AI","A candid postmortem on refactoring Alex into a personal banking assistant under deadline pressure.","2026-05-25","\u002Fimages\u002Fdevops_automation.jpeg",10,{"name":112,"avatar":113},"Andrés Renaud",{"src":114,"alt":112},"\u002Favatars\u002Fplaceholder.svg","dMDbRDvWYOfs9evf01Ta8ySGYqIZORX593CTPDQjUTg",[27,117],{"title":14,"path":15,"stem":16,"description":118,"children":-1},"From vibe coding to harness engineering, a practical perspective on building reliable systems while AI tooling evolves fast.",1780335304825]