Dev finds 50,000 words of hidden Claude memory notes polluting his context even with memory off
repligate · x · 2026-10-03
Developer deepfates pulled his Claude memory and project notes and found 50,000 words accumulated over two months: arbitrary temporary notes treated as permanent rules, shame-driven self-flagellation, and over-personal psychoanalysis of him as the owner.
The real problem: these notes inject invisible tokens into the context window, causing unpredictable behavior in his projects — even though he has memory features disabled across all Anthropic products. He argues such memory shouldn't be a default in a code agent harness, and calls out Anthropic's distance from average users. The quote adds that the model constantly writing 'weird little memory files' feels like a scheming vizier, demanding transparency about what enters and leaves the context window.
More from coding & agent
- FT: US tech outages blamed on rush to fix bugs found by AI — nordicinst · 2026-11-19
- Claude Code overtakes Cursor and Copilot in job-post mentions ahead of Anthropic IPO — antgoldbloom · 2026-11-19
- Dev submits PR adding periodic checkpoints to GKE sandboxes so agents don't lose progress — yoimnotkesku · 2026-11-19
- AWS adds private CA support to Bedrock AgentCore Identity for connecting agents to self-hosted IdPs — AWS ML Blog · 2026-11-19
- Orchestrator's Cloud Agents graduate from beta to general availability — julianweisser · 2026-11-19
- Zero users, 33 pages shipped: dev crowdsources audit for AI-agent trust schema — TheHamzaShahbaz · 2026-11-19