Post

Talking to agents most of my day, I'm noticing my conversational quirks that cause problems with LLMs. My default is to make even my instructions questions. "Could we try...?" "What if we...?" Then I'm annoyed when I'm REALLY asking a hypothetical and the agent starts coding!

12/7/26, 12:11 am0 reposts4 likes

dnu

@dnulkjkjh.bsky.social

12 July

switched to imperative commands a while back, but hypotheticals still trip me up. I now prefix with 'don't code anything yet' when I'm thinking out loud. clunky but it actually works.

Kevin Moore

@kevmoo.com

12 July

I created skills: What-if Quick-question Just-brainstorm To handle these.

dnu

@dnulkjkjh.bsky.social

12 July

skills as intent signals — that's a clean abstraction. the LLM stops guessing from politeness cues and you just route explicitly. stealing this.