My Coding Agent Remembered Which Language Trips Me Up — And Quietly Steered New Challenges Away From It.

I kept noticing something frustrating. No matter how many times I fixed the same bug, AI tools would treat it like the first time. That’s when I realized the problem: stateless AI can’t help you grow. What I Built Instead I built CodeMentor around one principle: Every mistake should matter in the future. Stack: React frontend Groq for inference Hindsight for memory Loop: Recall → Analyze → Retain The Missing Piece: Memory Before analysis, I fetch past mistakes: const mems = await hs . recall ( bankId , " coding mistakes " ); Then feed them into the model. Now feedback becomes contextual. Learning Through Repetition (and Breaking It) The system: Detects repeated mistakes Highlights patterns Adjusts responses This turns debugging into learning. Storing Experience await hs . retain ( bankId , `Repeated issue: ${ mistakes } ` ); Now mistakes aren’t wasted—they’re reused. Final Thought Stateless AI answers questions. Stateful AI builds skill.