Mainplane uses context retrieval instead of compacting AI-agent sessions

A post about Mainplane argues that compaction can lose details such as error messages and file paths. Its author reports that, in a roughly 400,000-token session using Opus 5.5, compaction took 38 seconds and cost $1.99 with an expired prompt cache, or 29 seconds and $0.17 with an unexpired cache. They report getting back to work through retrieval in 15 seconds for $0.10, using the same session and model.