Task Fidelity Scaling Laws — Kobie Crawdord, Snorkel
Snorkel argues task quality and data quality are the same, driving agentic training outcomes
“the quality of data is critical”
Snorkel developer advocate Kobie Crawford presented 'Task Fidelity Scaling Laws,' arguing that task quality and data quality are effectively the same and directly govern training outcomes for agentic, terminal-bench-style tasks. The talk extends Snorkel's longstanding data-centric thesis into the agentic space but offers framing more than concrete benchmark results. It matters as a signal that data-quality discipline is becoming central to agent training, though the excerpt lacks hard findings.