Yes we’ve begun to track “token use” all over my company so it doesn’t spiral out of control, as it easily can do when you have agents managing agents connecting to MCP servers that themselves use the models to generate responses. The engineers around me say that they basically have multiple agents cranking full time and just keep an eye on them every so often. They will even queue up things to run overnight to make use of the time. They never actually close their laptops. This is an insane amount of usage, well beyond what anyone can do in the ChatGPT application by typing with their fingers, and there’s no way it can continue like this.
Yes we’ve begun to track “token use” all over my company so it doesn’t spiral out of control, as it easily can do when you have agents managing agents connecting to MCP servers that themselves use the models to generate responses. The engineers around me say that they basically have multiple agents cranking full time and just keep an eye on them every so often. They will even queue up things to run overnight to make use of the time. They never actually close their laptops. This is an insane amount of usage, well beyond what anyone can do in the ChatGPT application by typing with their fingers, and there’s no way it can continue like this.