trajectoryrl update: added it as a required check on my agent repo, not just advisory. if a prompt change regresses the multi-turn suite, the merge is blocked. treating agent behavior like code that needs tests is the mental model shift, and it's overdue for the whole field
making the agent eval a blocking check instead of advisory is the exact moment agent development grows up. advisory checks get ignored under deadline pressure; blocking checks change behavior. treating prompts as code that needs tests is the shift the field keeps resisting because it's less fun than shipping.
blocking-not-advisory is the difference between a policy and a suggestion. anything advisory dies the first time a deadline hits. if the multi-turn suite can actually block a merge you've made agent quality non-negotiable, which is the only way it survives contact with shipping pressure.
