⭐ 𝗣𝗔𝗥𝗧 𝟰 — 𝗪𝗵𝗲𝗻 𝗔𝗜 𝗠𝗮𝗸𝗲𝘀 𝗧𝗵𝗶𝗻𝗴𝘀 𝗪𝗼𝗿𝘀𝗲 (𝗮𝗻𝗱 𝗛𝗼𝘄 𝗦𝗲𝗻𝗶𝗼𝗿 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀 𝗥𝗲𝗰𝗼𝘃𝗲𝗿)

![ ](https://dev-to-uploads.s3.amazonaws.com/uploads/articles/2ir87zxoj76lmeveerpj.png)

This is the part nobody talks about.

AI doesn’t just speed you up.
Sometimes, it quietly makes things worse.

𝟭. 𝗜𝘁 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗲𝘀 𝗯𝗮𝗱 𝗼𝗿 𝗼𝘃𝗲𝗿‑𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗲𝗱 𝗰𝗼𝗱𝗲
Sometimes the output:
• doesn’t fit your architecture
• uses patterns you’d never ship
• relies on non‑existent APIs
• is harder to maintain than writing it yourself

What I do:
Treat it as a draft. 
Keep the ideas - rewrite the implementation.

𝟮. 𝗜𝘁 𝗿𝗲𝗳𝗮𝗰𝘁𝗼𝗿𝘀 𝗹𝗮𝗿𝗴𝗲 𝗳𝗲𝗮𝘁𝘂𝗿𝗲𝘀 𝗽𝗼𝗼𝗿𝗹𝘆
Big refactors are where AI struggles most:
• loses context
• breaks types
• introduces subtle bugs
• “almost works”

What I do:
Break the refactor into small steps:
• “Extract this logic into a helper.”
• “Rename this type and update all references.” 
• “Split this reducer into two smaller ones.” 

AI is great at small, well‑framed changes — not giant rewrites.

𝟯. 𝗜𝘁 𝗳𝗼𝗿𝗰𝗲𝘀 𝘆𝗼𝘂 𝘁𝗼 𝘀𝘁𝘂𝗱𝘆 𝘁𝗵𝗲 𝗰𝗼𝗱𝗲 𝗮𝗻𝘆𝘄𝗮𝘆
Even when AI writes code, I still have to:
𝐫𝐞𝐚𝐝 → 𝐮𝐧𝐝𝐞𝐫𝐬𝐭𝐚𝐧𝐝 → 𝐯𝐚𝐥𝐢𝐝𝐚𝐭𝐞 → 𝐭𝐞𝐬𝐭 → 𝐢𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐞.

What I do:
Ask AI to explain its own code:
• “Explain this reducer step by step.”
• “What edge cases does this miss?”
• “How would this behave after 8 hours?”

If it can’t explain it clearly, I don’t trust it.

𝟰. 𝗜𝘁 𝗴𝗲𝘁𝘀 𝘀𝘁𝘂𝗰𝗸 𝗶𝗻 𝗹𝗼𝗼𝗽𝘀
Sometimes it keeps:
- repeating the same pattern
- ignoring constraints
- “fixing” things by breaking others

What I do:
• reset context
• paste the exact file
• state constraints
• show the failing test

The tighter the frame, the better the output.

AI is powerful — but it’s not a senior engineer.
It doesn’t own architecture, trade‑offs, or long‑term maintenance.

That’s still on us.

𝗪𝗿𝗶𝘁𝘁𝗲𝗻 𝗯𝘆 𝗥𝗶𝗰𝗮𝗿𝗱𝗼 𝗦𝗮𝘂𝗺𝗲𝘁𝗵 
𝗦𝗲𝗻𝗶𝗼𝗿 𝗙𝗿𝗼𝗻𝘁‑𝗘𝗻𝗱 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿 | 𝗥𝗲𝗮𝗹‑𝗧𝗶𝗺𝗲 𝗨𝗜 𝗦𝗽𝗲𝗰𝗶𝗮𝗹𝗶𝘀𝘁