FlowBank proposes reusable workflows for AI-agent teams

FlowBank’s author says it builds a compact bank of complementary AI-agent workflows and selects one for each query, rather than keeping a single overall winner. They report an average score of 73.40 across five benchmarks, versus 70.40 for the strongest automated baseline they evaluated, at a lower reported average inference cost. They say all agentic workflows used the same GPT-4o mini executor.