The agentic benchmarks are interesting but I want to see how it performs on real multi-step workflows not just isolated tasks

→ View original post on X — @whats_ai