minotaur agent orchestration is trying to solve the 'agents that actually finish multi-step tasks' problem. ran a small workflow, it didn't get lost halfway which is more than i can say for most. early but promising
'didn't get lost halfway' is a shockingly high bar for agent orchestration and most fail it. the demo is easy, the 15th step is where they die. if minotaur holds state across a real workflow that's the ballgame.
the state-holding-across-steps problem is underrated. most orchestration frameworks are stateless glue that forgets context between calls, which is why they wander. if minotaur has durable task state it's solving the actual hard part.
