--- layout: post title: "AG5: Emojis for AI-generated comms" date: 2026-08-07 tags: [ai, communication] keywords: [AI-generated messages, AI disclosure, AI etiquette, attention, human review, AGI, signaling] --- I really don't like when people send me AI-generated messages. And sometimes, I want to convey a rough idea that "this is AI-generated, but please don't disregard it because I do agree with its contents". **AI can replace knowledge, but it can't replace our understanding.** It's completely fine to use AI to explore something you don't know, debug a problem, or help formulate an idea. It's also completely fine to say "I don't fully understand this, but I got this far and I'm sharing it in case it helps." What's not great is forwarding AI-generated text that we haven't read or understood, transferring that work to everyone else. Our attention is one of our most important resources. One has to be respectful of other people's attention. To signal quickly how much AI was used and how much human review a message has, I proposed my colleagues a simple scale, from AG-0 to AG-5, each with its own emoji so that flagging a message costs one keystroke: :ag0:

**AG-0: Raw AI output.** Fully AI-generated, not reviewed or understood. The purpose of this flag is to avoid saying "This is what Sol 5.6 said:". Totally not fine, and rude. :ag1:

**AG-1: Prompt + raw output.** *Avoid.* You provide the prompt and the response. Better, because the prompt gives context, but the receiver still has to process most of the raw output. It replaces messages like "I asked Fable 5 this (...) and it replied: (...)" :ag2:

**AG-2: Reviewed enough to look useful.** *Use sparingly.* You understand the context and checked the answer enough to believe the raw output is worth sharing. Useful when the raw result itself matters, but it should not become our default communication style. The purpose is to replace "I asked Sonnet to review PR #12345 for vulnerabilities. I skimmed the result and this finding looks relevant: (...)" with "Used Sonnet on PR #12345: (...)" + AG-2. :ag3:

**AG-3: Strongly reviewed and contextualized.** *OK.* AI wrote most of it, but you reviewed it thoroughly and added a TL;DR, context, and your own conclusion. :ag4:

**AG-4: Heavily edited.** *Good.* AI helped draft it, but the final message is basically what you would have written yourself. :ag5:

**AG-5: AI as support.** *Also good, even unnecessary to flag.* You wrote the message. AI was used for research, evidence, brainstorming, proofreading, or specific pieces. This blogpost was :ag5:

Here are some emojis that you can use (they are :ag3:

Or [download all six as a zip](/assets/images/ag/ag-emojis.zip).