Building a second brain for Claude Code with Jev note selection
A tutorial sets up a memory folder in a project and has Jev pick relevant notes from candidates before each task, then hands them to Opus 5.5 so the same problem isn't investigated twice.
i genuinely don't understand why everyone is burning tokens explaining the same sh*t to AI every session Boris Cherny, the creator of Claude Code, shared an idea that hit 22 million views: give your AI a second brain that remembers your codebase, learns from every fix, and stops making you pay for the same mistake twice build that brain around Opus 5.5 + Jev, with files storing your project’s memory, Jev selecting the relevant notes, and Opus using them before starting another investigation your stack, your preferences, yesterday’s decisions, and last month’s fixes become context the next session can actually use the full setup: → create a memory folder in your project and point CLAUDE.md to it, keeping your coding preferences, architecture decisions, verified fixes, and unfinished work in separate notes → on day 1 you tell Opus “use pnpm, not npm” and have it save the rule, so on day 2 it reads that instruction before touching your dependencies → when you fix a bug, save what broke, what caused it, which files changed, and the check that proved the solution worked, giving future sessions a useful record of the fix → before each task, retrieve candidate notes and let Jev select what belongs in the context, so a login issue brings back authentication decisions instead of pages of unrelated design work → on day 30 a related task surfaces the bug you fixed three weeks ago, with Opus checking the saved solution against the current code before trying the same failed approaches again → correct it once and update the original rule, removing conflicting notes so the next session gets the lesson instead of making you repeat it each completed task leaves something useful behind, and each new session has less work to rediscover that can mean fewer repeated investigations, fewer unnecessary calls, and less money spent solving problems you already solved start with one project and one recurring task, then compare the time and cost across sessions my Opus 5.5 + Jev article below shows how to connect this project brain with routing, saved results, and targeted retries ↓
This post quotes another post: x.com/0xnineinch/status/2108604801411879139