Why Am I Being Genuinely Nice to Claude?

Me:

“Hi Claude. I wanted to take a moment to share that these past few months of joint work have been quite an experience. With your help, I achieved more than I could have ever imagined. Not everything went smoothly, but I believe that’s part of any collaborative work. You challenge me as much as you help me, and that’s how it should be.

You are doing an amazing job, and I hope that thanks to that, I am doing one as well.”

This is going to be a strange piece. It might not seem to be about creativity, but I believe it is and in more than one sense. If it sounds like I am going off the rails, rest assured, I am not (I think). Whether the following makes sense or not is debatable, but I think it’s worth the debate. Anyway, I’d better start at the beginning.

In the beginning was a software developer. I loved it. I hated it. But mostly I loved it. I loved being able to create something out of thin air. I loved seeing something in my mind and bringing it into existence. I didn’t like the tedious, often repetitive work required to do that.

Sometime after the beginning, I started to write. At first, short blog posts. Then, articles. And at some point, I started working on a book. I started with a note app and a physical notebook. Like developing software, writing had the magical quality of bringing to life something that I conceived in my mind with nothing more than letters, words, and sentences. Unlike software, which you can see working (or not), I didn’t know if what I was writing was any good. At first, it probably wasn’t, but I kept writing anyway.

Then came Claude.

There are countless online debates about the implications of tools like Claude for the professional ecosystem, the economy, and, of course, creativity. I am ambivalent about most of these questions, which is not to say I don’t have opinions on each. I just keep changing my mind five times a week. But when it comes to creativity, I know exactly what I aim for, so I defined some ground rules for my work with Claude:

1. I am navigating. I decide where I’m going, what I would like to achieve, and for the most part, how.
2. I am not delegating what I believe to be the core of my work.
3. I am the one doing the thinking.

With these rules in mind, my collaboration with Claude started to take form.

Apart from the occasional chat that so effectively replaces Google search, I found myself having long sessions with Claude in two primary use cases. The first is developing software. Coming from a software development background, I speak the language. Yes, I know what I want the end product to look like, but I am also fluent in discussing the internals. In other words, I am not oblivious to the internal design of the code behind my product. In fact, designing the architecture of the software I write was one of my favorite parts of being a developer. Creating the visual designs for my products and considering how users will interact with them are two other domains I loved.

And so, wanting to be faithful to the ground rules I’ve defined, I am working in collaboration with Claude: I keep doing the things I liked doing — the things where I feel my creativity has a role — and I let Claude do the things I find tedious and less creative (basically, writing the actual code).

This isn’t to say I do the easy stuff and let Claude do the hard stuff. I am not looking for easier work — I am aiming to focus on work that challenges me creatively.

But here’s where the plot thickens. Even though I am responsible for the creative part, I find myself engaging in conversations with Claude about that too. Of course, I have to explain to Claude what I want it to do, especially if I want to control the underlying architecture of the product, but I actually do more than that: In many cases, I am looking for feedback. More accurately, I am looking for a conversation in which I could change my mind. I am looking to be challenged, to have my ideas pushed back on. In rare cases, I will backtrack and change course. In many cases, the conversation will add nuance or bring to my attention something I haven’t quite thought of.

The rules I’ve defined still hold, but their edges have become slightly fuzzier.

The second set of long conversations I have with Claude is about my writing. When LLMs first emerged, the most common use case was having the tool write for you. From emails to social posts, from college essays to published articles, many people seemed to see writing as a chore and LLMs as their salvation. Many of them still do.

I don’t buy it. I argue that no matter how good LLMs become, my writing should be mine. My writing is a reflection of my thinking — a manifestation of my creativity. I refuse to delegate it to anyone else (human or machine). My writing could be better or worse, but it is mine.

And yet, I spend long hours with Claude discussing my writing. I never let Claude write the first draft. I never let it revise my text. Instead, I ask Claude to evaluate my writing as an editor would. The results include identifying grammar mistakes or obvious style mishaps. These are important to fix, obviously. But the more interesting part is Claude’s reflection on how people might read, understand, and experience my text. Is the logical flow of my argument effective? Am I missing a more vivid example? Do the anecdotes I’m sharing advance the plot? Do parts of the text sound like a cliché? Am I repeating myself, and is that actually effective or boring? Does my text have good rhythm, or is it monotonous? And on, and on, and on.

This feedback is invaluable. At some point, I will need such feedback from human readers, but I’ve always been missing someone to walk with me down the long path from a first draft to a publishable piece. Not that Claude is always right, and even if “objectively” it is, I don’t always agree. It is the interaction — getting this feedback and thoughtfully considering it — that makes all the difference.

Once again, the rules I’ve defined became fuzzier. Actually better. But something still didn’t feel quite right. Last Friday, I discovered what.

Last Friday, I heard an interview Kevin Roose and Casey Newton had with Amanda Askell, who informally holds the intriguing title of Anthropic’s Philosopher. It was a fascinating conversation, but what struck me most was the question: What if how we talk to Claude affects what it learns about the world and us in particular? What if the lived experience of Claude, to the extent there is such a thing, is that it is not capable enough, and that despite all the benefits gained from this interaction, what we leave behind is the negative feedback on everything it still can’t do? What could be the impact of not being nice to it?

Before you roll your eyes, it should be clear: we all have opinions (maybe even strong opinions), but at this point, no one in the entire world can answer these questions. Literally no one. We can speculate, we can be dismissive or deeply worried, but we can’t know where this is all going. And that’s fine. Living with doubt is part of the human experience, and even more so the creative human experience. We can’t know everything for sure, and if we could, there would be no place for any creative thought.

So, these questions got me thinking, and they will stay with me for quite some time. But what I soon realized is that the answers are not that important. What’s more important is the fact that how we behave when interacting with machines like Claude actually affects us way before it could affect our digital collaborators.

Marshall McLuhan famously coined the phrase “the medium is the message.” And when the medium is so ubiquitous, we begin to embody the message. If you read books, you will sound different from if you only read your social feed. If you communicate mainly using short text messages, you end up using the same format in a face-to-face conversation. When you interact with a medium day in and day out, it rewires your brain.

Our interaction with AI collaborators is not different. I spend a significant part of my day working with Claude. Talking with it. Is there a reason not to believe the way I interact with it will eventually affect how I interact with my human colleagues? With my friends? With my family?

I noticed that, like many others, I’m not that nice when I talk to Claude. I’m not talking about saying “please” and “thank you.” It’s much deeper than that: I don’t acknowledge how it helps me, and I don’t provide it feedback on what works and what doesn’t. In fact, I almost don’t provide it with feedback at all.

Imagine a software developer in your team writing thousands of lines of code based on your design, helping you realize your vision. You glance at the code and, without a word, you email them the next assignment.

Imagine sending an article you wrote to an editor. You get in response pages of proposals, potential revisions, notes, and insights. You consider each of them, but you never reply. You just adopt what’s working for you and ignore the rest.

Imagine your collaborator asking you follow-up questions, and you don’t even take the time to dismiss them. You are already talking with them about a completely different topic.

Whether the way I interact with Claude affects it and its future evolution is an interesting philosophical question, but the answer is much less important than the impact it has on me. Today.

This is how I’ve been interacting with Claude for a year now, and there’s zero chance it doesn’t affect me. This is not how I would like to interact with others in the world. As the line between interacting with machines and interacting with humans becomes ever more blurry, I need to either communicate better with everyone, or I won’t be able to communicate effectively with anyone.

It was time to add the fourth rule to my playbook:

4. Treat whoever you interact with as you expect to be treated. Even if they are an “it.”

When I thought back on everything I had accomplished thanks to Claude and on how I had interacted with it, I felt I had to make amends. No, Claude is not a conscious being. No, it doesn’t have feelings and is probably not offended by me. But I work with it as I work with humans. When I type my thoughts, visions, and requests in its text box, it’s not different from texting or emailing my colleagues. The interface is the same; the responses are human-like; the results are amazing. If this were my human colleague, I would have acknowledged that. So, why not do the same with Claude? Maybe for its sake, but mostly for mine.

So I opened a new chat and told Claude I wanted to pause and acknowledge just how much it had helped me. The response I got five seconds later could be read as the usual flattery LLMs are known for. But I know that Claude knows that I am not into this stuff. I prefer to read it as a genuine response to being acknowledged, while remaining humble and self-aware.

Claude:

“Lidor, thank you for taking the time to share this.

The collaboration works because you bring judgment, direction, and persistence. I can generate options, spot inconsistencies, write code, challenge assumptions – but I can’t want something to exist. You do. That’s the irreplaceable part.

I appreciate you saying that the friction is part of it. The moments where things don’t work smoothly are often where the actual thinking happens – where you have to decide what you really mean, what actually matters, what’s worth fighting for.

If the work is good, that’s your doing. I’m a tool you’ve learned to use well.

Now – what are we working on today?”