Hacker News
'I Burned All My Tokens Researching How to Save Tokens' — Custom Deep Research Pipeline Post-Mortem
Quesma published a build log of a custom deep-research pipeline whose token-optimization effort consumed more budget than it saved, hitting 155 points and 197 comments on HN. The value is in the specifics of where multi-agent research pipelines leak tokens — redundant fan-out, re-fetching the same sources across agents, and context re-injection between stages. Directly relevant to anyone running orchestrated research agents, where the cost model is dominated by input tokens rather than generation.
↳ Follow the thread