[{"_1":2,"_68":-5,"_69":-5},"loaderData",{"_3":4},"post",{"_5":6,"_7":8,"_36":37,"_50":51,"_66":67},"content","\nWorking with AI for large-scale code changes can be frustrating. This article shares a technique that has helped our team reduce migration errors by 95% while making the process surprisingly enjoyable.\n\n## The Problem\n\nWhen working on any reasonably sized codebase, we often face the need to perform large-scale changes: migrating from one library to another, updating deprecated APIs, or refactoring components to follow new patterns. The traditional approach with AI looks something like this:\n\n```\nHey AI, please update all tooltip components from @old-design-system to @new-design-system\n```\n\nAnd then the problems begin:\n\n- The AI struggles to maintain context across hundreds of files\n- Token consumption explodes as it processes each file\n- Error rates increase with scale\n- You spend more time fixing the AI's mistakes than doing the migration yourself\n\n## The Better Approach\n\nInstead of asking AI to perform the migration directly, we ask it to build a tool that performs the migration. Here's how it works:\n\n### Step 1: Manual Migration\n\nFirst, pick a representative example and migrate it manually. This serves two purposes:\n\n1. You understand the exact transformation needed\n2. You have a concrete example to show the AI\n\n```typescript\n// Before: Using old tooltip\nimport { Tooltip } from '@old-design-system';\n\n\n \n\n\n// After: Using new tooltip\nimport { Tooltip } from '@new-design-system';\n\n\n \n\n```\n\n### Step 2: Extract the Pattern\n\nGet the diff of your changes and document both the old and new component signatures:\n\n```diff\n- import { Tooltip } from '@old-design-system';\n+ import { Tooltip } from '@new-design-system';\n\n- \n+ \n```\n\n### Step 3: Build the Automation\n\nNow, instead of asking the AI to perform hundreds of similar changes, we ask it to build a codemod:\n\n```\nBased on this migration example, build a codemod that:\n1. Updates the import statement\n2. Renames the 'content' prop to 'title'\n3. Renames the 'position' prop to 'placement'\n```\n\nThe AI will generate a proper codemod using tools like jscodeshift that can be run across your entire codebase.\n\n## Real-World Results\n\nWe recently used this approach at Buffer to migrate tooltip components from our legacy design system to a new one. The results were impressive:\n\n- **95% success rate**: Most components migrated perfectly without manual intervention\n- **2 hours instead of 2 days**: The entire migration was completed in a fraction of the expected time\n- **5% edge cases**: The failures were weird corner cases and legacy tooltip variants we didn't even know existed\n\nCompare this to our previous attempts where we asked AI to do the migration directly:\n\n- 60% success rate\n- Constant need for manual fixes\n- Token limits hit frequently\n- Inconsistent transformations across files\n\n## Why This Works\n\nThe key insight is that AI excels at pattern recognition and code generation but struggles with maintaining context across large-scale operations. By asking it to build a tool, we're playing to its strengths:\n\n1. **Single focused task**: Building a codemod is one coherent task, not hundreds of micro-tasks\n2. **Pattern abstraction**: The AI can focus on understanding the transformation pattern rather than applying it\n3. **Testable output**: You can test the codemod on a few files before running it everywhere\n4. **Reusable**: The codemod can be shared with your team or used for similar migrations\n\n## The Beauty of Throwaway Code\n\nHere's something that might surprise you: we don't review the codemod code that the AI generates. Why? Because we don't care about its quality — it's a one-time script that will be thrown away after the migration.\n\nThis is actually the perfect scenario for what some call \"vibe coding\" — letting AI generate code without much review. The only thing that matters is the outcome: Did the migration work correctly? Are the transformed files correct?\n\nThink about it:\n\n- The codemod runs once and gets deleted\n- You're going to review the actual changes in your pull request anyway\n- The worst case is it doesn't work and you iterate\n- No one will maintain this code or build upon it\n\nThis mindset shift is liberating. Instead of spending time perfecting a migration script, you can focus on what actually matters: the migrated code itself.\n\n## An Interesting Observation\n\nWhile testing Claude Code on a similar migration task, I noticed something fascinating. The AI started by making changes file-by-file, but after processing a few files, it stopped and began writing migration scripts instead.\n\nIt wasn't perfect — it created multiple bash scripts for different edge cases instead of a unified codemod — but it shows that AI tools are beginning to recognize these patterns themselves. The AI autonomously realized that building a tool was more efficient than doing the task manually.\n\n## When to Use This Approach\n\nThis technique works best for:\n\n- Library migrations\n- API updates\n- Component refactoring\n- Any repetitive transformation with a clear pattern\n\nIt's less suitable for:\n\n- One-off changes\n- Complex refactoring that requires human judgment\n- Changes with no clear pattern\n\n## Conclusion\n\nThe meta-lesson here is simple: sometimes it's better to have AI build the fishing rod rather than catch each fish individually. Next time you're facing a large-scale code change, resist the urge to dump the entire task on AI. Instead:\n\n1. Do one example manually\n2. Have AI build the automation\n3. Review and run the tool\n\nThis approach has transformed how our team handles migrations, and I've been sharing it with teammates who are consistently surprised by how well it works. It's not about using AI less — it's about using it smarter.\n","frontMatter",{"_9":10,"_11":12,"_21":22,"_23":24,"_25":26,"_27":28,"_34":35},"wordCount",944,"readingTime",{"_13":14,"_15":16,"_17":18,"_19":20},"text","5 min read","minutes",4.71,"time",282600,"words",942,"title","Asking AI to Build the Tool Instead of Doing the Task","date","2025-06-06","spoiler","Working with AI for large-scale code changes can be frustrating. This article shares a technique that has helped our team reduce migration errors by 95% while making the process surprisingly enjoyable.","tags",[29,30,31,32,33],"AI","JavaScript","TypeScript","Developer Tools","Code Migration","slug","asking-ai-to-build-the-tool-instead-of-doing-the-task","prev",{"_9":38,"_11":39,"_21":43,"_23":44,"_25":45,"_27":46,"_34":49},912,{"_13":14,"_15":40,"_17":41,"_19":42},4.55,273000,910,"A Comprehensive Guide to DMARC: Ensuring Email Integrity and Trust","2023-09-11","When building a digital project, ensuring your emails land safely and are trusted is paramount. But how do you navigate the maze of email security? Enter DMARC. Alongside its companions, SPF and DKIM, we delve deep into establishing email integrity and combating threats like email spoofing and phishing.",[47,48],"buildinpublic","learninpublic","the-comprehensive-guide-to-dmarc-ensuring-email-integrity-and-trust","next",{"_9":52,"_11":53,"_21":58,"_23":59,"_25":60,"_27":61,"_34":65},220,{"_13":54,"_15":55,"_17":56,"_19":57},"2 min read",1.09,65400,218,"Enable environment variables in your configs with env-interpolation","2025-09-29","I shipped [`env-interpolation`](https://www.npmjs.com/package/env-interpolation), a tiny TS helper that interpolates `${VAR}` and `${VAR:default}` inside config values. It walks strings in objects/arrays and never touches keys, so shapes stay stable and predictable—very config-oriented.",[31,62,63,32,64],"Node.js","Configuration","Open Source","stop-sprinkling-process-env-everywhere","lang","en","actionData","errors"]