codebase-coffee vs asking the agent

A coding agent is useful when your intent is already clear: trace this call, explain this test, find the owner of this configuration, or compare two implementation options. You can follow the answer immediately with a narrower question.

That flexibility is the point. There is no reason to turn every urgent investigation into a curriculum session when a direct answer will unblock the work.

Ad-hoc Q&A starts with your current attention. It is less reliable at surfacing a quiet convention, an architectural seam, or a recurring gotcha elsewhere in the repository when none of those are blocking today’s task.

Without a memory layer, it is also easy to revisit familiar ground while important areas remain untouched. A useful answer in one chat does not automatically become a deliberate learning sequence.

codebase-coffee starts with a repository scan and writes a ranked curriculum to `.codebase-coffee.md`. Each normal session picks one concept per day and follows the same brief contract: what it is, why it exists, file pointers, one gotcha, and an invitation to go deeper.

The memory file marks served concepts so the habit moves through the codebase instead of depending on chat history. A teammate can also queue-jump the curriculum by leaving a note for a concept that deserves the next coffee.

Start with direct questions when work demands an answer now. Use codebase-coffee when you want a low-friction daily practice that chooses a useful concept before urgency chooses one for you.

A coffee can become a conversation: accept the optional walkthrough, ask follow-ups, or inspect the referenced files. See the full flow in [how it works](/codebase-coffee/how-it-works), then choose your host from the [install guide](/codebase-coffee/install).