1/ Two years running a multi-agent system in production taught me one thing I didn't expect:
the intelligence was never the hard part.
making stochastic components behave predictably enough to bill a customer for β that was.
9/ worth it in mid-2025? yes β buy options weren't ready and we shipped.
starting today? evaluate the durable-execution engines FIRST. the pattern is free; reimplementing it needs a real reason.
full write-up π