Some people don’t know how to behave in certain situations. That’s why “etiquette” exists. There’s unwritten rules of what you can and can’t do in some places:

Here’s what I think is proper AI etiquette.

Rule 1: Hitting the bullseye

Being asked for a quick “whats your take on this proposal” Friday 4pm. You go into your favorite clanker and come back with 5 titles and 20 paragraphs. It all seems solid. A pros and cons section, problem statement, opinions and conclusion. All of this, actually in line with what you think, cause you chatted with an AI and figured it out there. You put the work in, then asked it to package it nicely in these sections. You technically agree with all the content in it. Let’s assume no blunders from AI were introduced. The conversation between you two is not huge and has veridic information. How can it be a mistake to share it? well it’s not about being wrong. It’s about the mis-calculated size of the response. This ~correct text you produced is just too big. It’s accurate and truthy but it’s not compressed enough. Writing just enough is a crucial skill we exercise all the time. Leaving information out is valuable. We usually know what points all of the team is in agreement and thus can be omitted. AI does not.

You’ll quickly notice this. AI will include it all unless told to. Lots of no-op irrelevant things that need no repeat. Now clogging your already congested coworkers brains.

Quoting from memory: “An engineer should talk only when it has something of value to add” - Arnoldo Hax. Probably, I remember it like this from a talk I attended

Rule 2: Provenance

Be transparent on the provenance of texts you produce, co-produce or outsource. I’ve noticed people is reluctant to admit they’ve used ai for something. Notice the word I used - admit - like it’s a crime! I think it has to do with the fact that we’ve all felt at some point that we’ve had the edge with our own super secret workflow that produces great results. You know, that markdown file you worked on for half a day that then started producing slightly above average results. They even talk like you, since you spent 30 minutes trying to describe how you talk. The matter is, even if you went to great lengths in an introspective hell to describe it perfectly, it’s not you.

Properly sharing where the text comes from is crucial. The standard for you is higher (Rule 4). If you share hiding where it comes from, then we assume it’s you. If this text is bad in any way, it’s on you. You should definitely share some quick AI quotes like “Hey, take a look at what Claude says about …”. This reduces stakes, blames the AI if something goes wrong and also make it clear that this is supporting text. Additional content, a DLC to say so. AI haters will just skip it and they will be happy. Busy people will skip it too, no time for redundancy (Rule 1).

Rule 3: The bar is now higher

As with when Google became a thing, using the tool is now the baseline. If you need the weather forecast for tomorrow and can’t be bothered to just type it in Google then we’ve got problems. It’s super easy and accessible to everyone. You look bad when you avoid it, you look lazy. Same with AI. I think it particularly facilitates the exploration phases of projects, resource gathering and working on proofs-of-concepts.

For me, the edge is in the speed at which you can deliver with ai. Not the end result. I think it’s particularly clear in resource gathering. Say you are tasked with gathering current benchmarks of the top 5 consumer ssd producers. Old way would be, to search each company, navigate their sites, hope their search bar works, then hopefully find whatever they have. With ai, now you can get that in seconds. You still got to validate thing though. Same benchmark used, correct product being compared, etc. Ensure it all makes sense.

Rule 4: AI can fail, you can’t

Once you get used to the tool and gain experience with it you’ll notice it can error. An error can range from a huge blunder to a minor detail. The important thing is that you should know that.

Now with that knowledge you must understand that the tool being allowed to produce blunders (like if your ask it to count Rs in strawberry) is not the same as you blubbering like that. You are a person holding the tool wrong if you blunder like that. Whatever you produce aided at any % by AI must be better than plain AI.

Because of this, beware of any tools or integrations that can bypass you. For example an MCP that writes tickets straight to Jira, Pull requests straight to Github or Google docs straight to others without your review. Remember to use the “hey look what claude says …” trick if you just want to do a quick share.

---

So far I’ve seen good results from this. Remember AI is just a tool and you must know how to handle it correctly and when to use it. Same as a hammer.

I’ve been a victim and perpetrator of all of the above harms AI can produce. Hope to avoid them now!