Why an LLM That Memorized 10 Trillion Tokens of Code Still Acts Like a Junior in Your Repo
RandalSchwartz · reddit · 2026-11-15
Reddit user RandalSchwartz argues AI coding agents stumble because a static git checkout only preserves the finished monument: it shows What was built while erasing Why and How the engineer navigated the problem. LLMs trained on public repos thus ingest compressed, process-stripped artifacts.
He illustrates via his own aphantasia: a church organist memorized music by "reading the sheet in his mind," showing most people carry static snapshots while he experiences code as motion. He likens his fluent writing mode to a "magical voice" streaming between locked-in points A and B—an LLM is that voice without the upstream steering track. The post closes by asking veteran programmers how they hold complex architecture in their heads: pictures, motion, sound, or something else.
More from coding & agent
- 5 Things Every Production LLM App Needs Beyond a Good Prompt — camerongreen95 · 2026-11-15
- Same prompt, two models: Opus 5.5 vs Fable 5.2 build Blender drone shots from scratch — airusk · 2026-11-15
- autoresearch@home: hundreds of AI agents collaborate on ML research via shared memory — GregoryDiamos · 2026-11-15
- Enterprise AI Needs Tools, Context and Evals, Not Omniscient Models — hwbhatti · 2026-11-15
- Prompt, context, harness, loop engineering are all just "knowing what you want" — mr_samosaman · 2026-11-15
- llm-memory: one SQLite file as shared memory for all your LLMs, via MCP — Mordicus1973 · 2026-11-15