Did I write this, or did an LLM?
I had a conversation recently with a friend and colleague whom I deeply respect. Someone who’s read my blog posts and told me they enjoyed my writing. We were talking about my process, and I mentioned that I use an LLM as part of it.
Their response surprised me: “This really affects how I see your writing.”
The work hadn’t changed. My ideas hadn’t changed. The words that resonated with them before were the same words. But now, knowing an LLM was involved in my process, it felt different to them.
“It just feels like I liked the robot,” they said.
That reaction has been sitting with me. Not because I think they’re wrong, but because I’m not sure they’re entirely right either. And the more I think about it, the more I realize how many unanswered questions are hiding in that discomfort.
The questions we’re not asking
When we have strong reactions to AI-assisted work, what are we actually reacting to? Are our responses consistent across different contexts? And more importantly, do we understand enough about how these tools work to even know what questions we should be asking?
I don’t have all the answers. I’m not even sure I have the right questions yet. But I think we need to start asking them.
We’ve been here before (or have we?)
Every major writing tool has faced initial resistance. The typewriter was too mechanical. Word processors made writers lazy. Spell check would ruin our spelling. Grammar checkers would make us dependent on machines.
Each time, we eventually accepted these tools as just tools. The thinking, the ideas, the authorship remained with the human.
But maybe LLMs are genuinely different. They don’t just transcribe or check errors. They can generate coherent prose from scratch. That feels like a categorical leap, not just another incremental tool.
Or maybe it just feels that way because it’s new and we haven’t figured out how to think about it yet.
I honestly don’t know. And I think that uncertainty is important to sit with.
The code question
In tech, using AI to help write code has become somewhat accepted. Developers use GitHub Copilot or ChatGPT to generate code, review the output, modify it, and ship it. Generally, no one questions whether they “really” wrote that code.
But suggest using an LLM to help organize your writing? Very different reaction.
Why the double standard?
Maybe it’s because code has objective success criteria. Does it work? Is it efficient? We can test these things. Writing is subjective. Voice, authenticity, originality - these aren’t testable. They’re based on something harder to pin down.
Or maybe we should be more concerned about AI-assisted code too. Maybe the acceptance in the tech world isn’t thoughtful adoption but premature normalization. Are junior developers who rely heavily on AI actually learning the fundamentals they need? I don’t know, but it’s worth asking.
What are we actually valuing?
Here’s a fundamental question I keep coming back to: Do the results matter, or does how we achieve them matter more?
If a piece of code works perfectly, solves the problem elegantly, and is well-architected - is it less valuable because someone used AI assistance to write it? Most people would say no. Working code is working code.
But writing feels different. There seems to be a stronger sense that the process itself has moral weight.
Why? Maybe because we see writing as more personal, more human, more connected to identity. Or maybe because we don’t fully understand what we lose when we outsource parts of the creative process.
And maybe it depends on what kind of writing we’re talking about. A novel, where prose is part of the art itself? That feels different from a technical blog post where the goal is clarity and communication.
But I’m not certain about any of this. These are genuine questions, not rhetorical ones.
What “writing” actually is
Let’s break down what goes into creating a piece of writing:
- Generating the ideas and thesis
- Developing supporting arguments
- Organizing chaos into structure
- Ensuring voice and accuracy
- Making editorial judgment calls
If someone does four out of five of those things themselves, and uses an LLM to help with organization, did they write it?
Maybe organization isn’t just one mechanical step. Maybe the way you structure an argument IS the argument. Where you put emphasis, how you sequence ideas, what you juxtapose - these might not be just presentation choices. They might be fundamental to meaning.
If that’s true, then “helping with organization” is actually helping with core intellectual work. And that changes the equation significantly.
But if the final output accurately communicates what you intended to communicate, does it matter how you got there? Or is effective communication the point, regardless of process?
I genuinely don’t know. Both positions seem to have merit.
The editor question
When someone uses an LLM in their writing process, is it functioning as an editor or as something else?
A human editor helps organize thoughts, suggests better structure, points out where things are unclear. They help you communicate more effectively. And we don’t think working with an editor makes writing less authentic.
So why would using an LLM to do similar work be different?
Maybe because we understand human editors. We know they’re helping, not replacing. We see their tracked changes and comments. We have back-and-forth dialogue about their suggestions. We have professional norms about their role.
With LLMs, we don’t have that clarity. The process is opaque. The norms don’t exist yet. The boundaries feel unclear.
Or maybe the difference is more fundamental: A human editor suggests changes to what you wrote. An LLM generates new prose based on your ideas. That might be a meaningful distinction. Or it might not be. I’m not sure.
Not all AI assistance is the same
Something I think often gets lost in these debates is that there are fundamentally different ways to use an LLM, and they produce fundamentally different results.
Compare these approaches:
Approach 1: “AI, write me a blog post about the future of device management.”
Approach 2: “Here are my thoughts about where device management is heading. I think X because of Y and Z. I’m concerned about A but excited about B. Help me organize these ideas into a blog post.”
The first is asking AI to generate ideas and content. The second is bringing your own ideas and asking for help with structure and clarity.
Then there’s the iterative process: taking that initial output, refining it, having back-and-forth dialogue about whether it captures what you meant, making changes, questioning word choices, restructuring sections. This collaborative refinement looks very different from one-shot generation.
Are these the same thing? Should they be evaluated the same way?
I don’t think so. But when we say “AI-assisted writing,” we often collapse these very different processes into one category.
It’s worth noting that low-effort content has always existed. Before LLMs, people churned out generic blog posts by following templates and copying competitors. Spam content farms have been around for decades. LLMs make it easier to produce this kind of content at scale, which is a genuine concern. But the existence of low-effort AI content doesn’t mean all AI-assisted work is low-effort.
The tool is neutral. How you use it matters. But we don’t seem to have good language yet for distinguishing between different modes of use.
My process (and what it reveals)
I’m neurodivergent. My brain doesn’t work in straight lines. My thinking is abstract and systemic - I see the whole interconnected web at once, not a sequence of logical steps.
I’ve always struggled to translate what’s in my head into linear prose that other people can parse. It’s not that I can’t think clearly. It’s that the way I think clearly doesn’t match the way neurotypical communication works.
So I use an LLM to help bridge that gap. I come to it with my thesis and arguments already formed. It helps me organize the chaos into something that follows neurotypical writing conventions.
But that’s not where the process ends. What comes next looks a lot like working with a human editor. I review what the LLM produces. I question whether it captures what I meant. I ask for revisions. We go back and forth - sometimes multiple times - refining structure, adjusting emphasis, clarifying points that feel muddled. I cut sections that don’t work. I rephrase things that don’t sound like me or don’t say exactly what I’m trying to say.
This iterative process continues until the final product communicates the message I’m trying to communicate. And here’s the important part: I don’t publish anything unless I stand behind every single word of it. If something doesn’t represent my actual thinking, it gets changed or cut. The LLM might help me organize my thoughts, but I’m responsible for what gets published.
For me, this is assistive technology. It helps me express ideas I already have in a format other people can understand.
But here’s the question this raises: Where’s the line between assistive technology and outsourcing intellectual work?
Speech-to-text helps people with motor disabilities, but it’s translating their actual speech. An LLM that “organizes thoughts” is doing something more than translation. It’s generating new prose that I then refine through collaboration.
Is that the next logical evolution of assistive technology? Or is it a categorical shift? Maybe it’s both?
I think it’s an important question, and I don’t think the answer is obvious.
The transparency problem
Should creators disclose when they’ve used AI?
If I use an LLM to help organize my thoughts, should I disclose that? If so, should I also disclose when I use Grammarly? Spell check? When I’ve worked with an editor? When I’ve talked through ideas with a friend who helped me clarify my thinking?
What level of tool assistance requires disclosure?
I don’t have a good answer to this. There’s genuine gray area here.
The assumption behind demanding disclosure seems to be that AI assistance makes work less valuable or less authentic. But if you can’t tell the difference without being told, what exactly is being devalued?
On the other hand, if there’s nothing problematic about using LLMs this way, why resist disclosure? Maybe transparency is valuable even when the assistance is legitimate.
I think we need better norms around this. But we can’t develop those norms without engaging with the technology and understanding what it does.
What the MacAdmin community reveals
I’ve noticed something specific to the MacAdmin community: there’s a particularly strong knee-jerk reaction that anything AI automatically equals bad. I see it more here than in other tech spaces I’m part of.
Maybe it’s about craft. Our community values deep technical knowledge and hands-on expertise. There’s pride in understanding systems at a fundamental level. AI tools can feel like they’re short-circuiting that understanding.
Maybe it’s about control. We work in a field where knowing exactly what’s happening under the hood matters. AI is opaque in ways that make us uncomfortable - and maybe reasonably so.
Maybe it’s about fear. Not necessarily fear of being replaced, but fear of what it means for human identity and value if AI can help with things we’ve always considered fundamentally human. That’s not an unreasonable fear.
Maybe it’s just new and we don’t trust it yet. We’ve seen enough tech hype cycles to be skeptical. That skepticism has served us well before.
But here’s what I find interesting: people who have that strong reaction will still tell me they love my blog posts. The content resonates. The ideas are useful. The voice feels authentic.
Until they find out an LLM touched it somewhere in the process. Then suddenly it feels less real, even though nothing about the actual content changed.
What does that tell us? That we’re reacting to the idea of AI involvement more than to any actual quality difference? Or that there’s something we value about purely human creation that we can’t quite articulate?
Both might be true.
What we might be losing (or not)
I don’t have all the answers here. Maybe there ARE important skills that atrophy when we rely on AI for organization. Maybe the process of struggling with prose does refine thinking in ways I’m missing when I use an LLM. Maybe I should be more concerned about this than I am.
The Luddites weren’t entirely wrong to worry about industrialization’s impact on workers and communities, even if the transformation was ultimately inevitable. Tools change us. They change what skills we develop, what we value, how we think.
AI won’t just make writing faster. It will transform what “writing” means. Is that good or bad? I don’t know. No one does yet.
But I think refusing to engage with the technology out of fear or principle means we won’t have a voice in shaping how that transformation happens. We’ll just be left behind, complaining about change we didn’t help direct.
The experiment
I’m not going to tell you which parts of this post involved an LLM. Does that bother you? Does it make you want to analyze every sentence, looking for tells? Does it make you trust this post less?
If you can’t tell the difference, what exactly are you reacting to? The work itself? Or just the idea that AI might have been involved?
The ideas in this post are mine. The arguments are mine. The uncertainty is mine. I stand by every word of it, and I’m genuinely grappling with these questions, not just defending my choices.
So did I write this? Or did an LLM?
Maybe the better question is: what do we need to understand about how these tools work before we can answer that question meaningfully?
Where this leaves us
I think we’re asking the wrong question when we argue about whether AI-assisted work is “real” writing or “authentic” creation.
The right questions might be:
- What are we actually valuing in different kinds of creative work?
- Does the purpose or goal of the work change what we value about it?
- Where are the meaningful boundaries between assistance and replacement?
- What skills and practices do we want to preserve, and why?
- How do we develop norms around disclosure and attribution?
- What does it mean for accessibility when AI can help bridge cognitive differences?
These are complicated questions, and I don’t have the answers. Neither do you.
And I think the only way we’re going to answer them is by engaging thoughtfully with the technology, not by rejecting it outright. That doesn’t mean uncritically embracing AI. It means being willing to use it, understand it, question it, and develop informed opinions based on actual experience rather than abstract fear.
I might be wrong about parts of this. You might read this and think I’m too cavalier about real concerns. That’s fair. These are genuine questions, and I’m genuinely uncertain about many of them.
But I think the uncertainty is valuable. I think sitting with the complexity is important. And I think we need to move past knee-jerk reactions - in either direction - and start having more nuanced conversations about what we’re actually worried about and what we actually value.