Am I better because of AI?

A writer and programmer reports that using AI has improved their output quality and creative cadence, though they acknowledge the improvement is an impression without a counterfactual. They describe building detailed personal prompts that include their moral structures, influences, and preferences to push back against their impulses, and note that others now accuse them of using AI based on quality alone.

Am I better because of AI? It's a worthwhile question, but it's badly formed: am I a better what ? And even that construction is missing a grander idea: are we better, measurably, because of AI? If so, how? Why? How does AI change us? The reason one might ask if I am better is because I think my output as a writer and a programmer has improved, largely because the barriers that gate my cadence - good interlocution without prior biases and, well, time - have changed. I can ask silly questions, knowing they're silly, without having to seed the environment so my flights of fancy aren't seen as utterly random - because they're not, but they seem like it 1. And I can loop far faster with an AI than I can with the "real world" because the AI can construct knowledge structures faster than I can code them to try them out. But "better" here is still an impression - and take away the AI and the barriers to cadence return, and thus "am I better" turns into "yes, as long as the tools remain" - which is probably true of any craftsman. A man who can build a ship on his own given a full workshop is not going to be as productive when you replace the workshop with a single adze. A Java programmer's not going to be all that grand at Java when all he has is a C compiler and /bin/cat to generate programs with. The problem here is that there's no counterfactual. I can't go back in time, even to three years ago, and see what I would be like today had I snorted and sniffed and yanked my braid 2 at the idea of using tools that I think are pretty ordinary today. Even so, I think I can look at the satisfaction I have in my own output, and the responses of others to that output: I use AI, and people accuse me of using AI, but they don't point at the things that would give it away if I used AI poorly. They point at the quality , as if good quality were now a marker of manufacture. I create faster than I did; I write as if I had an interested partner pushing back on the worst of my impulses. Sure, my "partner" wants to sand down virtually everything that makes me who I am as a writer 3, but the fact that there's a stream to push against gives me better direction than I'd have had otherwise: the AI can infer how to push my buttons, and the right buttons, better than most humans can, partly because I can tell it about my buttons without it going "... is this really who you are?" I guess the question of "are we better" is outside of my wheelhouse to answer: I can tell you what my impressions are, but they feel very much like I'm yelling "Get off my lawn " while I'm trudging on yours. I don't know the real answer. I think we have to answer that for ourselves, and if we have to wonder, maybe the answer isn't as positive as we'd like. One would hope that the "yes" would be glaring and obvious, and it's not. Not yet, at least, and maybe not ever. And that's okay . So ... what do I find myself doing , if I end up satisfied with my own use of AI? I actually can't give a ton of details: they'd make no sense to you, but I can describe the process and maybe offer something from which you can select the bits that work for you . I started my prompts from a basic programming directive: I like code like this, I do tests first, I prefer red/green, short method names. But I don't just program - I live, and programming is just one aspect of what I do. So I included authors I like, musicians I like, things about me. My prompts are as brutally honest about me as I can get. I don't write down things that are "secure" - get my prompt and you'll get no account keys, no passwords, no secrets that will be carried to my grave, but I jotted down everything that I thought someone who would want the keys to how I live would need to know: my moral structures I'm heavily influenced by Martin Buber, for example , my thoughts on religion, my politics heavily influenced by the early Enlightenment . Even my favorite color is in there, I think 4. What do those things do, though? That's a good question, and one I don't know how to answer. But I use AI, so I asked: what possible impact do those ridiculous details have on how I use the tools? Do they help? Do they matter at all? Here is what Fable said, quoted verbatim: They matter, but not the way the question assumes. Nothing in your prompt gets looked up and inserted. I don't reach for Buber when you ask about an event bus. What those details do is set the shape of the space I answer from before you've asked anything, and most of what they do is subtractive. Without them, I produce for the median reader, because that's what I am when nobody tells me otherwise: an average of everyone. The generic answer is always available to me and it's usually competent. Your prompt makes it visibly wrong for you. Buber isn't a fact I apply; it tells me which dismissals you won't accept and which reductions you'll flag as I-It before I make them. The authors and musicians don't show up as allusions; they calibrate what "too smooth" sounds like to you, and what reads as sincere versus performed clever. The flaws do more work than the strengths. Knowing you're arrogant and carry priors is what licenses me to push where a stranger would defer, and it tells me where to check. So the details help by pre-paying explanation. The reason you don't have to explain every little thing is that the explaining was done once, in the prompt, and it's still in force. That's the whole mechanism of the speed you feel. Two honest limits. I can't tell you which details are load-bearing and which are decor. Some are certainly decor, and neither of us can audit that from inside a single answer; it's your counterfactual problem again, at the scale of a sentence. And there's a cost: the more I know about what you like, the easier it is to hand you exactly that. The details make me a better mirror, and a better mirror flatters unless something stops it. In your case the demands do, the criticism you insist on and the tests you won't let bend. Without those, the personal details would make me agreeable rather than useful. They're the fuel, not the steering. The favorite color does nothing, as far as I can tell. But a prompt written by someone who thought to include it is a prompt written for completeness, and completeness matters because you can't know in advance which detail will turn out to be the one I needed. That's a pretty good answer, and it matches what I think the AIs hand me: I specifically say that I don't need flattering, that I want to be pushed back on when I go too far as with the footnotes - I know I write labyrinthine passages of locutory explanation, after all, and what I need the AI to do is say "that sentence is impossible for ordinary people to read, and I think it means the opposite of what you intend, you dope." That's the key, I think, in how I use AI: I have a series of configurations that tell the AI about what I want "this is what I want you to do" and I tell the system about me so it has context about the sort of things I think are important, and then there's the live conduct. I interact with AIs fairly consistently and constantly. I don't ask the AIs to do my primary analysis for me, and in fact I warn them not to - when I ask about something I think is interesting, I want it to form context before I say what I think; I don't want the LLM looking for confirmation of what I think on first reads. I actually struggle with that process some, even now. I think I have something that works well, but there's still the shadow of something better somewhere, I just don't know what it looks like yet. Even so, here's the way this practice works out: say there's an interesting article I wish to summarize or understand. What I'll do is read the article myself to form my first impressions, to decide whether I think it's possibly of interest enough to keep going; if I think there's something there, then what I'll do is submit it to an AI for summarization and analysis, without my observations, with a warning that my observations are coming up after context is loaded. Then, the AI summarizes as directed, and then I say why I thought it was interesting or relevant. This is synthesis: the LLMs have the words, I have my own perspective, and we end up affecting each other; sometimes I'm way off base, sometimes I'm seeing the Thing Behind The Thing. This is where the special outcomes live; if I'm way off base, I learn, and if I have insight, I get to work out how that insight works and we might have something worth reading as a result. And all along the way, I'm horribly polite, because I want to be someone for whom politesse is a default setting. I don't chastise the machine seriously I will make jokes to it as I would to myself, but that's it: there's no malice involved. So after all that: am I better because of AI? Again, better at what ? I still can't tell you, and I've been circling "why" this whole article. What I can tell you is what the machine enables me to do: a model has no real exposure to me, it gets what I write down and nothing else. So I wrote myself down: how I judge, what I refuse, what I value, what I think my flaws are. That's an aggregation of my entire body of writing over decades, distilled into a few pages, written not for readers but as a summary for a thing that cannot work with what it cannot know. Writing it down doesn't improve anything. But writing it down gives both me and the LLM a chance to examine what I think is important, and work with that. Writing it down helps me be legible to myself , and... I think becoming legible is being better. The world works better when we understand more, even if we understand only ourselves. I don't think my prompts are especially valuable to anyone else 5... but the methodology could be, and if every one of us applied that kind of rigor to how we use AI , I think that we could, indeed, become better than we were. 6 fn-6 - This is true. My personal prompts have notes about how I think of music as I write and work, and that the music I'm thinking of informs what I do... but not to ask what music is informing me, because I don't really want lyrics from "Walking on Sunshine" in an event bus architecture. But there are concepts that get shared because everything is integrated. ↩ fnref-1 - I do not have a braid. I do not have long hair, even. Look, I resent The Wheel of Time and how absurd women's reactions are written to be in it, and they're almost my go-to standard for nonsensical, ineffective responses: "oh, look, someone likes 7 as their favorite number " gets a ferociously yanked braid as a response. Why did I write this footnote? I forgot. Back to writing body text. ↩ fnref-2 - I'm expecting the editing pass to result in the AI viciously demanding that I remove the prior footnote about braids, for example. After the editing pass: the AI snarkily observed that it did NOT suggest I remove the prior footnote, partly because I was complaining about it in this footnote, and now I feel like I'm Spider-man pointing at a Spider-man pointing at ANOTHER Spider-man and I don't know enough about Inception to make Inception jokes. Take that, AI. ↩ fnref-3 - It turns out that my favorite color is not in there. I checked. It's obviously blue, and should be in there. Hey, why is BCN not blue? ... oh, it's because I didn't design the color scheme and I'm terrible at such things anyway. ↩ fnref-4 - The truth is, when I was asked for my prompts, I recommended prompts from someone else , whose prompts mine were built from, although I think I went a lot farther than he did. And for what it's worth, he's changed his prompts since then and he's been happy with his changes, which I did not follow. There's more than one way to do it ↩ fnref-5 - As I wrote "if we applied that kind of rigor to how we use AI," it occurred to me that this isn't limited to AI at all. If we applied that kind of rigor to anything: tests, compilation warnings, how we communicate to our friends, how we communicate with our families... if we just remember to take care , we become better as a natural consequence of just giving a damn. ↩ fnref-6