Ask HN: Do you thank your agents when they did a good job?
Sometimes I catch myself doing this and it feels pretty silly in retrospect
Or sometimes they fix a bug and I instinctively go back to tell them the fix worked even though I could just close the session at that point. Should I thank my washing machine for washing clothes? I'm careful not to project humanity onto an inanimate objects. I never write thanks specially if we are finishing. If there are no more interactions is simply wasting money. But when adding new messages I almost always start with "ok great, now let's continue with..." when previous job is ok, so I "reinforce" the clanker as much as I reinforce myself in my train of thought. I remember reading something from OpenAI’s CEO once where he said they churn millions on people saying please and thank you. All the time. Sometimes they do an absolutely bang-up job. I do, and frankly I enjoy my day more when I don't feel like I'm being a jerk, even towards a robot. But mostly as part of a follow up message. I also think it signals that I was happy with that solution in the ongoing conversation. On the other hand, I'm much more sloppy with my writing. The token cost is irrelevant, tool calls and reasoning probably dominates. The model's reward function is its own reward. No need to reinvent the wheel. Rather not waste tokens. Saying thank you literally costs you money. Only for the models where I'm not paying per token. You soon won't be able to afford to thank if they start decommissioning old models unless you run local. You are literally wasting energy, it's like leaving your care running for a few minutes for no reason. Not so much "thank you" but I find myself telling it stuff like "could you please do x etc" a lot. I've come to terms with it. If AGI ever comes it'll go like hey, at least this sack of meat said please.. Let's kill it last :-) Me: Thanks! Claude, you did a great job. Claude Code: Anytime. I'm glad we nailed it together. Happy to help. bla bla bla.... Me: Thanks! Codex, you did a great job. Codex: Understood. It finally clicked to me when Peter Steinberger mentioned that 50% of Codex's workforce is European. Can't relate. ChatGPT is so much more snivelling than Gemini, for example. I dont, but sometimes I say please. Best regards