Developers complain that AI repeats the same mistakes. "I've explained five times that we don't use Redux here." "It suggested FTP again instead of a queue." "It keeps ignoring our naming conventions."<\/p>\n
That's not a model problem. It's an engineering gap.<\/p>\n
<\/p>\n
Why the agent "doesn't learn"<\/h2>\n
Every new agent session starts from a blank slate. The model has no memory of previous conversations. Everything it knows about your project is what you put in its context. If you put nothing specific in \u2014 you get the average of the internet.<\/p>\n
Without domain context, the agent falls back on its training distribution, where production-grade architectural decisions appear far less often than beginner tutorials. The result: FTP instead of event-driven, global state where local state belongs, camelCase where your codebase has used snake_case for three years.<\/p>\n
The problem isn't that AI "doesn't learn." The problem is that no one gave it anything to learn from.<\/p>\n
The mechanism: two files<\/h2>\n
Any agent that automatically pulls a rules file into context at the start of each session can become an accumulating system with two lines of instruction.<\/p>\n
For Claude Code, this is CLAUDE.md<\/code>. For Cursor, it's .cursorrules<\/code> or .mdc<\/code> files. The principle is the same: instructions stored in the repository get loaded automatically on every run.<\/p>\n
Add two lines to your rules file:<\/p>\n
- If you made a decision that proved correct and cost real effort \u2014 log it in decisions.md (pattern + why it worked, 3 sentences max).\n- If you made a mistake that caused a problem \u2014 identify the root cause (not the symptom) and log it in mistakes.md.<\/code><\/pre>\n
That's it. Two plain text files at the root of the project.<\/p>\n
From the first session, the agent starts filling both files. Every subsequent session opens with them in context. The agent now sees that in this project:<\/p>\n
\n
dual write is forbidden (and why),<\/li>\n
Redis cache is the wrong tool for this data type,<\/li>\n
last time this pattern was used, it caused this specific failure.<\/li>\n<\/ul>\n
Mistakes stop repeating. Domain context is in the toolchain now \u2014 it wasn't before.<\/p>\n