Principles we should follow when using AI (or at least, principles I am following)
Contents
This article is non-technical. I might present my personal workflow of working with AI Agents and other tools in the future, but this here is about everyday AI usage.
We as in humanity have developed certain etiquettes in how we are using technology. As Yousr’Allah Allouani notes, it is considered disrespectful checking your phone for new messages while sitting in a Cafe with a friend. Or checking the time all too often. And we need the same for AI. I’m unsure as to whether “Etiquette” is the right word for that here, but this is how I believe you should use the current state of this technology, and a “rulebook” that I am subconsciously following.
This entire page is to be taken with a grain of salt. Life isn’t always that serious, and at times we just want to have fun, so then please ignore anything I am saying here. But there were plenty serious situations where I wished people got some inspiration from this list.
My Rulebook
- Own the result.
- Don’t lose your standards and vision.
- Keep thinking, stay curious.
- Don’t send or present me direct AI output.
- Understand the technology.
- Trusting ChatGPT more than a human expert.
- Keep words simple.
- Saying AI couldn’t do something.
- Use the technology.
- Avoid void productivity.
1. Own the result.
What I would argue is the most important thing, is that you own what you share, in the sense that you are proud to stand for the result, as if you had created it yourself. When you’re writing a text, sending me some information, prepare a presentation or code a product, you are responsible for the result and outcome. It has your name on it, and that doesn’t change even if you disclose using AI. This closes up on this Hacker News discussion, but I simply can’t stand „Gemini gave me that“ as an excuse for not being in charge when things go sideways. Or when your PR (for the non techies, change to code in a project), that is signed off by you, has a mistake, you own it. You are responsible. It’s not „oh, Claude made that mistake“, you did.
Mistakes happen, that’s completely normal. And then you fix things, but you are still responsible for the mistake. You can’t blame ChatGPT, that’s just the tool you used. Similarly, I am by no means saying that you shouldn’t be using AI (see point 9). I am saying is that we shouldn’t act as if it was one of us.
2. Your standards and vision
On a slightly different note, another thing that would fall under AI etiquette, is that you need to hold high to your standards. AI slop is a huge problem. Let me explain.
Doing 80% of the work takes 20% of the effort and time, it’s the last 20% that is hard. Keep pushing, until you have something that is on your standards. Don’t stop before that just because you have something that ticks the boxes. To succeed here, you need to have a very strong vision in advance, which is the hardest part here.
This goes for so many categories.
I have no problem with people using AI for writing. In fact, for certain things, you should and I do. Personally, I like writing texts like these myself, it’s a certain craft that just feels beautiful (see point 3), but if you don’t like writing and still have thoughts to share, absolutely, please, use AI to get those thoughts down to paper (or the screen, for that sense). But make sure it is your thoughts come out, and that it reads the way you intended. Don’t stop when the text you have is “somewhat like you imagined it” or when it reads plausible.
I have no problem with people using AI for coding. In fact, you should, for everything. You can create something that ticks the spec-sheet faster than ever, but all of a sudden it’s literally only ticking the box, and simply because it didn’t go though your personal craft and care, it doesn’t, feel, right. And it is so hard to define, what feeling right even means. Here, once again, you need to keep your vision in your mind, and work towards that, instead of towards ticking that box. This is what is causing the downfall of Apple’s software. SWE’s at Apple are using Claude Code everywhere, have never been properly trained, and while on paper all boxes are ticked, it feels like nothing is actually working, things are buggy, animations are broken, and it simply isn’t pleasant anymore to use.
3. Keep thinking, stay curious
Now that machines can think, it is easier than ever to offload all of your thinking to the machine. We have yet to see where this will lead us, but (besides the societal perspective), you need to think whether or not you are personally comfortable with that.
We can’t unlearn how to think, how to trust our intuition, how to express ourselves, etc. We can’t just rely on a neural network to make all these decisions for us, because where does that leave us? The next consequence is that we stop caring for things, because everything seems being done and we can just lay back. You do not want to live that way.
An unspoken rule that I am trying to live by is to create (write, talk, cooking food, draw, …) more than I consume (watch socials, listen to music, …). This doesn’t work out most days, but it is an interesting goalpost in my opinion. The main point here is, that you should keep caring and thinking. Use AI as assistance, for both input and output, but stay the entity that is genuinely interested.
4. Don’t send me AI output
This is closely related to point 2 on the list. What I mean by “AI output” is one-shotted first-try zero-effort AI copy-pasted content.
I hate reading a text or listening to a speech that is obviously AI generated. I immediately hate that email you sent me, as soon as I get the feeling it is AI generated. I hate that speech or presentation that you are trying to present, if it’s so obviously AI generated and I feel like you don’t actually know what you’re talking about. I hate that Linkedin post that is absolute slop.
Because I get the feeling you don’t care. It feels like it’s not your opinion, but you’re just the outlet. I feel like it’s not you speaking on stage, but you’re just the outlet presenting ChatGPT’s opinion. This might or might not be the case, but that feeling alone is enough to keep me from caring about whatever you’re presenting or telling me in that email.
Nobody will be mad if your email isn’t impeccable. In fact, I’ll be happy when there is a typo, and I believe this is a wonderful shift in mindset that we have ahead of us.
5. Understand the technology
What I am not saying is that you should understand how a transformer works, or how an Agent works, or anything like that. You don’t need to understand these details if you only want to use the technology.
But I want you to understand the minimum about the technology in its current form here, I’ll try to make this as non-technical as possible:
- It returns text to answer your question. Everything that it can do apart from that, such as searching things on the internet or taking your DoorDash order has been deliberately built in by humans.
- It does not “know” facts in the human sense. It does not have a database of truths or a conceptual understanding of the world; rather, it recreates information that appeared frequently or reliably in its training data, and can decide to set out and do an internet search. Information you get can be wrong.
- It does not have feelings or consciousness. Any expression of emotion, personality, or “desire” is an imitation of human language patterns found in its training sets, programmed to make the interaction feel more natural.
- It only knows what it is given, or what it can find out with the tool it is given. When you ask Gemini a question like “am I on the Pro plan”, this doesn’t work unless someone at Google decided to code this feature.
6. Trusting ChatGPT more than a human expert
A thinking system that can use the internet sounds like an entity that you should trust more than humans. And there is a thin line between using ChatGPT for getting information to question things given by other people (which is great!) and just ignoring what other people say because a beloved ChatGPT says otherwise.
There are really discomforting stories of Slack chats where developers decide to implement a feature a certain way and a PM just sends a screenshot of an ChatGPT answer that says how it should be done differently, and then that PM insists of doing it the ChatGPT way.
Humans have instincts and taste. And I understand that I am moving on very thin ice here, but I would argue that the combination of instinct and taste is often so much more valuable than sheer knowledge. Sometimes you are right, even though everyone is telling you that you are wrong. Sometimes the human expert is right, even though all information is telling you he is wrong. A perspective.
Take the information from that AI output and ask the human in question what he thinks of this. Maybe it’s a side he hasn’t explored yet. Maybe it’s absolutely correct. But don’t assume it is right just because ChatGPT gave it to you.
7. Keep it simple
The beautiful thing about words is that there are hundred ways to express yourself, to say the same thing. Following up on point 4, we as humanity have started valuing more complex expressions higher than simpler, straight forward variations. It used to mean that more care has been put in the result.
Today, generally, the opposite is more likely to hold. Text is cheaper than ever, but the cost of content hasn’t moved. On an unrelated note, this is the same way code is cheaper than ever, but the cost of engineering hasn’t moved.
Generally, the simpler you can put something, the better. This is an old principle in the world of teaching and education. It is harder to explain the topics in simpler terms than with a lengthy, complex wording. The same applies to all other texts. If you’re building a product and you describe it as an “AI powered Hybrid RAG platform with in-browser intelligence and local ONNX embeddings“, you lost me. If you’re asking for something through an email and it takes you 100 words what could be reduced to 10 words with a question mark, take that route b, please.
A complex formulation is no longer a sign for having put in more care.
8. Saying AI couldn’t do something
You couldn’t manage using AI for that.
This once again ties to point 1, saying that it is you doing the work, and you are using AI as a tool. Inherently AI doesn’t “do” anything without you guiding it what to do, and when we say “AI couldn’t do something”, what we actually mean is that I couldn’t do it, even with the help of AI. That’s a formulation that is much more truthful.
9. Use the technology
The more you use our current state of AI technology, the more ways you come up with on how you could use it. Every person who is using ChatGPT for the first time has completely underestimated this technology before trying it. Every person who is using Claude Code for the first time has completely underestimated this technology (agents) before trying it. And only after trying it, people are genuinely impressed and start using it more and more, and come up with more and more use cases.
There have not been a lot of technological inventions where this flow works this way.
I want to encourage you to use this technology to the fullest. I ended my presentation about agentic coding by saying that you should always be pushing the limits. If you get one-shotted everything you throw at Claude Code for example, then the conclusion can’t be “oh, well now my work is chill”, it should be questioning on how you can do more, to keep pushing the limits of what’s possible.
10. Avoid void productivity
On a different note, realize when what you are doing with AI is actually slowing you down. Coming back to the example of writing an Email – going back and forth with ChatGPT 10 times to get to your desired tone of an Email (congrats! you’re following point 2), just consider that you might be faster bringing your thoughts down to paper directly.