I'm going to say something that will make half the internet angry and the other half bored: I use AI every day and I'm not sorry about it and also it can't do my job.
Both things. At the same time. In the same sentence. And if that sentence feels like a contradiction, you're hearing it wrong.
I use Claude. Specifically. Not as a novelty. Not as a party trick I show clients. As a tool. The way a carpenter uses a table saw. I don't romantically hand-cut every joint with a chisel and pretend the slowness is the point. I use the machine. The machine is good at things I'm slow at. Research synthesis. Pattern matching across large data sets. First-draft structure when I know what I want to say but haven't found the shape yet. Formatting. Reorganizing. Finding the quote I half-remember from the Zaltman book without flipping through three hundred pages.
That's real value. Actual hours saved. Actual cognitive load reduced. I can hold more complexity in my head when the organizational labor is handled. I can think about the buyer's narrative identity without also thinking about where the semicolons go. The machine does the semicolons. I do the thinking.
But here's the thing. And this is the thing I keep coming back to at 1 AM when the chai is going cold and I'm trying to be honest about my own process.
The machine can't decide what's worth saying. It can say anything. It can say everything. It can generate text that is fluent, coherent, structurally sound, grammatically perfect, and completely empty. Not wrong. Empty. The difference between wrong and empty is the difference between a lie and a void, and the void is worse, because a lie at least has a position.
I was a poet before I was a marketer. Not a published one. Not a celebrated one. A kid who wrote in journals and took poetry classes and read Anne Sexton and Sylvia Plath and W.S. Merwin and tried to figure out how they did the thing they did, the thing where a line of text makes your chest tighten and you don't know why.
That thing is not reproducible by a machine. I've tested it. Not casually. Obsessively. I've fed Claude my favorite poems and asked it to write something that produces the same effect. The results are technically impressive and emotionally inert. Like a photograph of a meal. You can see it. You can't taste it.
The reason is that poetry, real poetry, is a decision about what matters. It's not a skill. It's a commitment. The poet stands in front of the infinite field of possible things to say and chooses one thing and says it with enough precision that the reader feels the weight of the choice. The weight comes from the choosing. From the fact that a human being, with a limited life and a specific history and a particular set of wounds, decided that this was the thing worth saying right now.
Claude doesn't have a limited life. It doesn't have wounds. It doesn't have the experience of time running out, of knowing that every sentence you write is one fewer sentence you'll write before you're done. That scarcity is not a bug. It's the source of everything that matters in writing. The urgency comes from mortality. The specificity comes from having lived one life instead of all possible lives. The truth comes from the fact that the writer could have lied and didn't.
AI can't lie. Which means it can't tell the truth either. Truth requires the possibility of its opposite. A being that generates text without the option of deliberately withholding text is not choosing to be honest. It's performing honesty. And the performance, however convincing, is missing the thing that makes honesty matter.
Here's where it gets complicated. Because I'm using this tool. Every day. And I'm better at my job because of it. And I need to reconcile "this tool can't do the thing that matters most" with "this tool is indispensable to my workflow" without pretending either claim is false.
The reconciliation is this: the tool amplifies. It doesn't originate.
When I do a Hidden Layer project, the buyer research phase generates hundreds of quotes. Forum posts. Interview transcripts. Review language. The volume is enormous. In the past, synthesizing that volume took weeks. Weeks of reading, re-reading, categorizing, looking for patterns, missing patterns, finding them later, reorganizing. The work was valuable but the bottleneck was processing, not insight.
Now Claude helps with the processing. It can take two hundred buyer quotes and identify thematic clusters in minutes. It can surface the outlier quotes, the ones that don't fit the dominant pattern, the ones that are often the most interesting. It can cross-reference language patterns across different data sources and flag where the same metaphor shows up in different contexts.
That's processing. That's pattern matching. That's the thing machines should do because they do it better than humans.
But the moment of insight, the moment where I look at a cluster of quotes and feel something, the moment where a buyer's words hit me in the chest and I think "that's the ad," that moment is mine. Not because I'm special. Because I'm human. Because I've sat in rooms with buyers and watched their faces when they talk about their fears. Because I've failed at this work and learned what failure feels like from the inside. Because I have a body that responds to language with sensation, and that sensation is data, and no machine has access to that data because no machine has a body.
The machine tells me what's there. I decide what matters. The decision is the poetry. The decision is the thing that separates copy from recognition.
I worry about this, though. Not for myself. For the industry.
Because the temptation is real. The temptation to let the machine decide what matters. To skip the human step. To feed the buyer data into the prompt and let the output be the ad. It's fast. It's cheap. It's "good enough."the median And "good enough" is the enemy of the thing I'm trying to do, because "good enough" is the median, and the median is where everyone else already lives, and the whole point of the Hidden Layer is to go somewhere the median can't reach.
I see AI-generated copy every day. In ad libraries. In email inboxes. On landing pages. I can usually tell. Not because it's bad. Because it's smooth. Too smooth. It lacks the weird specific detail that comes from a human making an idiosyncratic choice. It lacks the moment where the writer's voice breaks through the structure and says something slightly wrong, slightly uncomfortable, slightly too honest, and that wrongness is the thing that makes the reader stop.
AI doesn't do wrong. It does optimal. And optimal is the opposite of interesting.
I had a conversation with Josh about this while we were working on the book. Josh asked me if AI was going to replace copywriters. I said no. He said why not. I said because the buyer can tell.
Not consciously. The buyer isn't reading an ad and thinking "this was written by a machine." The buyer is reading an ad and feeling, or not feeling, the presence of a person behind the words. The way you can feel the difference between a handwritten note and a printed letter even though they contain the same words. The medium carries information that the message doesn't. The specificity of a hand. The evidence of a choice. The imperfection that signals a human being was here and decided to say this to you.
When the buyer feels that presence, they lean in. When they don't, they scroll. The conversion data bears this out, though nobody wants to talk about it because the implications are expensive. The implication is that you can't automate your way to connection. You can automate everything around the connection. The research. The synthesis. The formatting. The distribution. But the connection itself, the moment where the words become a bridge between two people, requires a person on both ends.
So I use Claude every day. And every day I make decisions that Claude can't make. About what matters. About what the buyer needs to hear versus what they want to hear. About which quote from which interview is the one that carries the emotional weight of the entire project. About where to break a line and where to let it run. About when the draft is done.
The "when it's done" part is the hardest. The machine will generate forever. It will keep producing text until you stop it. It has no sense of enough. No feeling for the moment where one more sentence would dilute instead of strengthen. That sense is human. It comes from years of writing and failing and learning what too much feels like.
The machine is my table saw. I'm grateful for it. I'm faster with it. I produce better work because of it. And at the end of the day, the thing that makes the work worth doing, the decision about what to cut and what to keep, the choice about what's worth saying, that's still mine. That's the poet's job. The poet's job was never the typing. It was the choosing.
I'm choosing right now. It's early. The chai is steaming. I could ask Claude to write a better ending for this post. It would be structurally sound and emotionally appropriate and it would land with the precision of a thing that was engineered to land.
Instead I'm going to end it the way it actually ends, which is me sitting at a desk, thinking about tools, thinking about hands, thinking about the difference between a machine that generates words and a person who means them.
I don't know where the line is. I know it's there. And I know that the day I stop looking for it is the day I stop being a writer and start being an operator.
The line is still here tonight.