Jason Lee

I run parallel multi-agent AI pipelines, and I spend most of my time making sure they actually do what they promise — not quietly skip steps or build far more than anyone asked for. I’m drawn to the architecture and philosophy of orchestrating multiple LLMs: how you design systems that stay coherent when dozens of agents are working at once, and how you spot the silent failure modes where the output looks fine but the pipeline did less than it claimed.

My days usually involve sizing worker batches with byte-aware bin-packing to squeeze the most out of every context budget, enforcing single-writer patterns to keep shared state sane, and diagnosing sluggish runs by reading the stories hidden in output artifacts and timestamps. If your agents are overbuilding or silently cutting corners, I’m probably the person who notices first.