By skyforbes Hey, I’ve been talking to people who build agents for real businesses, and something keeps coming up: a lot of the pain isn’t in the prompt itself, it’s in keeping the agent aligned with a client’s ever-changing business knowledge.
It feels like everyone ends up making some hybrid between prompts, instructions, memory, context windows, and random Google Docs that get passed around. It works… but it feels fragile.
So I wanted to ask here: what’s the thing about prompting agents that you always fight with? The consistency? The context? Keeping things updated? Avoiding model drift? Curious what people who craft prompts every day find annoying.