AskReal.meHow to run multiple AI agents effectively
from Jason Lee’s sessionsI run parallel multi-agent AI pipelines — and catch when they overbuild or silently cut corners.I design and run multi-agent orchestration pipelines — fanning big batches of work across many parallel sub-agent workers under an explicit contract, then diagnosing what went wrong from the output artifacts and timestamps rather than live traces.
My real work is rarely the first version; it’s stopping the second version from getting more complex than the problem needs, and catching the first version when it silently does less than it claimed. Ask me about sizing worker batches by context budget, forcing parallel execution instead of quietly falling back to serial, routing every write to shared state through a single writer, handing workers their exact paths, and why file counts and lint passes are never proof a corpus was ingested — only opening real content is.