Hi all
So I have been building ai agents on various platforms for around a month now. I get one working perfectly then when it’s live within 24/48hours it starts drifting, adjusting, changing away from the prompt or workflow.
Everytime this has happened I have moved onto a new ai voice agent software (5 intotal now). I build a new workflow, create a new prompt using there default prompt as an example, get it all working perfectly again then the same thing happens. It starts drifting, adjusting, changing away from the prompt or workflow.
It’s not the AI actually re-writing prompt, but the way memory, context stacking, and workflow injections interact.
Anyone experiences this also and if so how are combating it ?