Agentic test processes, LLM benchmarks, and other notes on agentic coding fr
Summary
Dan Luu outlines experiments with agentic coding, LLM benchmarks, and testing workflows, highlighting high variance across models and tasks, problematic false positives, and the limits of relying on benchmarks alone. He discusses caveman mode, agentic loops, and the need for systematic evals, data-driven test generation, and diverse personas to keep agentic processes productive.