Definition
Engineering memory is the structured, reusable, and improvable working record of how engineering knowledge is created, validated, preserved, generalized, applied, and repaired by humans, tools, and AI agents.
Working Claim
Memory is not raw capture. It is promotion.
Engineering work leaves behind commits, reviews, logs, incidents, documents, scripts, and session traces. Those artifacts matter, but they only become memory when the lesson has an owner, a boundary, a reuse path, and evidence that can return to the owner when later work changes the lesson.
The source has to be available where the next work runs, within the limits that environment can state honestly. A trace can recover the question; it cannot replace the current source that answers it.
Use
Use this concept to distinguish durable engineering knowledge from temporary execution residue.
Open Questions
- Which lessons deserve promotion into repository documentation, workflows, skills, models, or articles?
- What evidence is enough to make a lesson reusable?
- How should AI-assisted work preserve useful patterns without creating competing sources of truth?
- How can a system show that its local working sources are current enough for a bounded task?