John Mikhail’s Post

Over the last 8 months I built a handful of codebases from scratch with varying complexity using Claude Code. It always started off great, then the quality degraded as the codebase grew bigger. Recently, I discovered that my agent had been quietly learning from our conversations for weeks using an experimental feature called Auto Memory. That took me down a rabbit hole into the different memory systems available to coding agents. I summarized what I learned and how I ended up using skills as my primary memory system. Would love to hear how you're managing your agent's memory. There's no agreed-upon best practice yet and we're all figuring it out together. https://lnkd.in/eYu3YUue