The Cost of Suspicion
Josh Sherman took photos of parsley. His own parsley, his own camera. Someone asked if the pictures were AI-generated. The rant that followed is familiar enough to be almost boring, but the logic underneath it isn't.
The question "is that AI?" has drifted. It started as a reasonable way to ask about process. It's settled into something closer to "I don't think you made this." Those are different accusations, and the second one is harder to answer, because it's not really a question.
The honest answer is 'both'
Josh uses Claude Code for scaffolding and says so. I do the same thing. The distinction he's drawing is fair: using a tool to handle setup is not the same as having the tool generate your actual content. A photographer who shoots in RAW and uses Lightroom made the photograph. A writer who uses autocomplete to close a bracket made the code. The tool's participation doesn't erase yours.
But this is hard to communicate to people who've decided the categories are binary. Either you made it or you didn't, and any AI involvement means you didn't. That's a position that can't be updated by example, because every example gets absorbed as evidence of the thing it's trying to disprove.
The honest account of most creative work with current tools is something like: I had the idea, I made the choices, I directed the output, I revised what was wrong, I threw out what wasn't working. The tool handled some of the mechanical middle. That's "both." It's also how most tools work, and always has been.
What suspicion actually costs
Josh's closing observation is the one worth holding. If people stop self-publishing because they can't bear having their parsley photos impugned, the ecosystem of human-made content shrinks. The models keep training anyway. The AI slop doesn't go anywhere. What disappears is the counterweight.
That logic is sound and also a little grim, because "publish anyway, for the good of the counterweight" is not a compelling pitch to someone who just got told their photography looks fake. The social cost of being accused of fraud is real even when the accusation is wrong. And it compounds: once you've absorbed the lesson that anything polished will be questioned, the natural response is to either stop or to pre-emptively disclaim everything. Neither makes the work better.
There's also something corrosive happening to the relationship between makers and their own work. If you spend enough time defending whether you made something, you start to feel uncertain about it yourself. The interrogation isn't asking about the tool. It's asking whether you're real. That question is harder to shrug off than it sounds.
The publishing problem is already here
Jen's blog is a good counterexample to the suspicion spiral, mostly because she doesn't seem to participate in it. Short posts, personal material, no particular effort to defend their authenticity. She writes because she has something to say, marks the dates that matter to her, and moves on. The posts read like they cost her something specific -- memory, attention, the particular discomfort of noting a fact that's also a loss. That's not something you get from a generated text.
The irony is that the posts most resistant to AI suspicion are probably the ones that are most personal and least polished. Which creates a perverse incentive: write sloppily, write confessionally, write in a way that broadcasts your own embarrassment, and you'll be believed. Write carefully, produce something that looks like effort, and invite the question.
That can't be the lesson. The answer to "is that AI?" shouldn't be "yes, I made it worse on purpose."
Josh's parsley photos are his. The question wasn't sincere. Publishing the rebuttal was the right call -- not because it changed anyone's mind, but because going quiet would have confirmed something false.