How to speak to AI agents effectively: key differences from human communication

A developer on DEV Community has outlined why dictating instructions to an AI agent requires a fundamentally different approach than speaking to a human or recording notes for yourself. Unlike a human listener, an AI agent treats every spoken word as a direct instruction with no one reviewing or interpreting it in between, meaning offhand remarks and implied constraints carry the same weight as explicit commands. The author identifies three core pitfalls: burying constraints as asides instead of stating them upfront, using multiple names for the same thing, and relying on spoken intonation to convey structure or emphasis — none of which survive transcription. They also caution against overloading a single session with multiple subjects, noting that even well-formed instructions get lost when too many decisions compete in one conversation. The practical advice is to state constraints before goals, use consistent terminology, number steps explicitly out loud, and type any literal details like file paths or commands rather than speaking them.