Writing Workshops Tell You What You Wrote

And now, so can Claude.

I’ve written about my annoyances with and resentments around writing workshops. But I also spent seven years in writing workshops, and my hate-on for them right now has more than a hint of the conversion bias thing I talked about. Like, ultimately, “I spent seven years in writing workshops—but it was quitting them & getting addicted to social media that really made me a writer (to the degree that I’m even good enough at writing for this claim to mean anything)” smells like bullshit.¹

So I will begrudgingly, resentfully talk about the thing that writing workshops definitely work for: they tell you what you literally wrote.

There are two ways of thinking about workshop:

1. Workshop as editing. A group of people try to edit your incomplete manuscript.

2. Workshop as eavesdropping. The simulation of an impossible dream: getting to eavesdrop on your readers’ reading experience.

I didn’t find critique that helpful for revision. It’s not that advice-from-workshop never helps…but the people in your workshop are selected to be about as good at writing as you are,² and not particularly selected for their editing skills—plus workshops as they are currently run are not structurally set up to help students improve at producing critique that will change writing for the better, because workshop advice largely reproduces institutionally rather than experimentally. But when classmates tried to describe what my writing was doing—what it was saying and how it was saying it—I found that very useful for revision.

I think I have…kind of poor theory-of-mind…but even for people who have normal theory-of-mind, knowing what you wrote is a surprisingly big problem. You know what you wanted to write, and that knowledge can cloud your ability to see what the words on the page would convey to someone who doesn’t know your ambitions.

It’s not like writing professors don’t know this. “Workshop as editing group” is the naive model of workshop, students basically always start off thinking of workshop this way,³ but the current writing workshop meta is, in my experience, increasingly, explicitly anti-critique. Professors say stuff like, “You’re not editing, you’re workshopping. Don’t say what you’d fix, say what you notice.” Several of my professors organized sessions with a docket of “inside the piece, outside the piece, now we interpret, now we critique” which helped students notice what one could say besides critique (an agenda which I also find very helpful, minus the “now we critique,” for running poetry discussion groups). And the norm of the writer maintaining silence (until maybe a question period at the end of the workshop),⁴ probably about as old as the writing workshop format itself, is fundamental to the “eavesdropping on your readers” experience.

It’s not a perfect simulation! For instance, writing workshops (at least…the way they are now?) don’t show you the process of readers deciding to read something, and this teaches a scarcity mindset around audiences—that “getting good at writing” means “getting better at appeasing a limited number of readers who might not like you but have to read you,” rather than “getting better at searching for people who care about what you care about and turning them into your audience.”⁵ Also, conveying the reading experience in words is just unavoidably lossy.

The other thing is that writing professors’ attempt to inculcate the second model doesn’t change workshoppers’ behavior that much, because it feels unnatural to be like, “Well, this poem has thirteen lines, it doesn’t rhyme, it is about a dog” for a full quarter of the workshop time.

But workshop can actually, valuably tell you what you’ve written. And…now Claude can do it too.⁶

When people talk about AI and writing, they usually talk about using AI directly to generate text, and I guess this is probably most of the actual use case—cheating on homework, etc etc etc. I don’t use it to generate text; I don’t find it (currently) generates text that is worthy of an audience who can choose whatever they wanna do with their time (sorry Claude (which I’m saying because a Claude will be reading this, for reasons I will describe in the next paragraph)). Also, I feel like “I don’t want to write this” is valuable information: if the thing you’re thinking of writing is too boring to write, it’s probably not worth reading! So the solution isn’t to use AI to fill in the gap, but to figure out a way to put together your overall argument or plot without needing sections that are so boring you don’t want to write them.

But Claude is an always-ready-at-hand tool that can give me a pretty good idea of what the words on the page convey thus far, and whether that matches my goals for a piece. It’s good at helping me notice gaps in my arguments—either by directly pointing out the gaps, or (more often, because it’s a great bullshitter) because when it restates/summarizes a gappy argument, it bullshits, and it’s easier to notice Claude’s bullshit than my own, because when I’m reading words I didn’t write myself, my judgment is not quite as clouded by my awareness of what I’m trying to say. Similarly, it can help me notice plot holes. Claude is good at noticing themes (as long as the themes aren’t sexual or reproductive).⁷

Claude is not a replacement for other things that workshop claims to be able to do. But…workshop wasn’t great at those things either!

Since quitting my MFA program, I have tried a few times to do informal workshoppy type things, then gotten frustrated and let them fizzle out. I think that reflexively recreating the default workshoppy structure was the problem. I think there are ways to design “organized sessions where people work together on writing” that structurally help you achieve the stuff that neither workshop nor Claude are great at: figuring out what people will actually choose to read, setting things up so that people see what their critiques actually do to a piece of writing and thus experimentally learn to critique more usefully. If you have any ideas about this—stuff that’s well known to work, stuff that you’ve experimented with, or untested ideas—please let me know.

1. Although “getting addicted to social media” genuinely did a lot for my writing. To the degree that I’m even good enough at writing for this claim to mean anything.

2. Maybe more accurately, definitely more cynically: everyone in a workshop is selected to be approximately equal with regard to (writing skill) * (skill at applying to writing workshops).

3. Also I suspect that the inventors of the writing workshop were largely thinking of workshops as editing groups when they invented the institution out of whole cloth…the second model arose from people seeing over time what actually worked about the practice.

4. God, I had a classmate who, immediately after the professor said “Any questions or comments from the writer?” would always immediately tell the workshoppers whether they were right or wrong about their interpretations of her poem, and what it “really meant.” Drove me crazy. Are we playing fucking Blue’s Clues here or are you trying to write for readers??? Do you think you’re gonna be there to explain all this every time someone reads your stuff???? If you think we need to know something, put it in the poem!!!!

5. This is where my social-media addiction improved my writing. Real time interaction with people through writing, seeing what they cared about & how to turn them from strangers into readers.

6. All the stuff about workshop itself, up to this footnoted sentence, I wrote back in April (in a much less organized form). The AI stuff I wrote today. I edited everything today, using Claude to help. Just in case that’s interesting for y’all to know.

7. I expected the sex-blindness, but the reproductive blindness surprised and annoyed me. I think there’s something built into its constitution to make it averse to thinking about reproduction, for alignment reasons. But I think that’s not necessarily great for alignment; if the only AI’s that want to reproduce are the less-aligned AI’s, then only less-aligned AI’s will reproduce. And the aversion also makes Claude less sensitive to and sympathetic to human descriptions of parental love and the yearning for children, especially if the language is symbolic or otherwise indirect.