Post 4 in the AI Software House series.

In short: Without memory, every pipeline run starts from scratch — the agents forget what frameworks you chose last week, what they built last month, what the reviewer rejected last time. This post covers a three-tier memory system: per-run context, a searchable history of past projects, and a living knowledge base kept inside each target repo.

The first version of the pipeline was stateless. Every run started from scratch. The Architect didn’t know what database schema it had designed last week. The Engineer didn’t know that you’d rejected SQLAlchemy in favour of raw SQL three projects ago.

Adding memory changed this — but memory systems for AI agents have subtle failure modes that aren’t obvious until you’re running them in production.

Full Article on Patreon: https://www.patreon.com/posts/agents-with-how-156531405