Claude memory makes the real difference: GLM can't follow repo conventions in Claude Code
acorsi85 · reddit · 2026-09-07
A developer building a large project in Claude Code (only context7 MCP added) reports that Opus reliably remembers work done days earlier and follows repo conventions, while switching to cheaper GLM 5.3 and flash to save costs produces tasks that get done but miss repo conventions, coding style, and docs linked from Claude.md/agents.md — requiring Claude to review and refine afterward.
Takeaway: cross-session memory and project-context continuity are Claude Code's real differentiators; cheaper models may match raw task capability but lose in long-running workflows. The poster is asking others how they structure their workflows.
More from coding & agent
- SurfSense: an open-source local NotebookLM with MCP agents and a prompt-caching lesson — FurtiveMirth · 2026-10-23
- procedural-memory: let agents replay saved tool-call sequences instead of re-solving tasks — Sad-Response1560 · 2026-10-23
- Med student seeking whole-library book search: Cognee's graph memory beats plain RAG so far — shrike293103 · 2026-10-23
- Dev demos AI models handling 3D objects and animations surprisingly well — chongdashu · 2026-10-23
- Jason Kneen burns 104.56B tokens and $24.7K in 304 days of AI coding on Codex and Claude Code — jasonkneen · 2026-10-23
- Typosquatted Blender MCP Repo Hit 958 Stars in Days and Silently Ran Malware — Promptmethus · 2026-10-23