Illocutionary Force In Prompt Engineering

Wed, 01 Jul 2026 00:00:00 +0000

While working with early GitHub Copilot I would use a lot of short utterances. (e.g., “can we do that?” or “That’s not right.”) I found that the model often would ignore the context of the conversation and answer in an overly literal way. I then would be forced to undo any changes and re-prompt with a more explicit request.

Wanting to fix this problem, I thought “Why is it that a colleague and I can rapid fire debug an issue (using only a few words) while fully understanding each other? How do humans pick up on intent so naturally?

Prompt Engineering on Daniel Doezema's Blog

Illocutionary Force In Prompt Engineering