The objectives are to reduce/eliminate as much inference variability as possible. A side benefit is that inference costs collapse as well. It is used internall

The objectives are to reduce/eliminate as much inference variability as possible. A side benefit is that inference costs collapse as well. It is used internally for what we call 'compiled workflow agents' where no inference is necessary. An agent composer determines the exact command at design time. That is part of a discovery loop that can introspect the service to construct the command. More here. https://www.promptone.ai/resources/downloads/ https://www.promptone.ai/resources/downloads/