One more post, before I lock in for my thesis…
The following post started on Bluesky, after Charlie Stross reposted a screenshot of AI nonsense from Threads:
https://bsky.app/profile/cstross.bsky.social/post/3msd75w7ka2xk
I’ve edited it slightly and reposted it here. There is a larger post in the works, titled “No-one Wants AI, Except For Them”… however, I had to remind myself that I am an MFA student in the final month of his thesis. I’m well past squeaky bum time – now I need all the time I can get to make sure my thesis is as polished as it can be.
In the meantime, you may consider the following to be a very brief, very provisional manifesto on how I think AI could be used in writing, from an individual point of view.
Spoiler (not a spoiler): AI is plugins.
I want to know what you think about this framing. I think it allows us to talk of AI technologies as being entirely optional, AND to show others what cognitive tasks we value. Plugins can help smooth out tedious, admin tasks; they can’t do your research for you, and they especially cannot write your story. My intent is not to do a bit of both-sidesing, but to start exploring how writers might attain centaurhood, while continuing to be critical of Silicon Valley and its products.
Just one more thing: I write these posts for free, and as long as I have a stable income, free is how these posts will stay. I personally don’t like the idea of paywalling posts that you might find useful. Nevertheless, if you do enjoy what you read here, and leave this post with a greater understanding of the craft, please consider leaving a small tip on Ko-fi:
Vibechecking “AI is plugins” (permalink)
For a few weeks I have been writing a little AI piece of my own. I would’ve liked to have gotten it out… a few weeks ago. Since then, both Charlie and Cory Doctorow have put out some nuanced posts about their views on AI:
http://www.antipope.org/charlie/blog-static/2026/08/on-the-non-use-of-ai-in-my-wri.html
https://pluralistic.net/2026/07/24/supplemental-income/
Ask any fiction writer on Bsky/Mastodon, and we will be quite happy to tell you where and why we think AI shouldn’t get involved in the writing process.
To the average “AI is inevitable2” pundit, to a first approximation, all writers are anti-AI in any and all cases. They have something of a point (although we do not have to hear it from them): it is much easier to show solidarity with writers you don’t know by showing anti-AI sentiment in any and all cases. The bagmen have so utterly poisoned the well, that any mention of AI with a positive connotation might suggest that we’d invite our adversaries to dinner.
I know. I know that you know that it’s coming. That 3-letter b-word I’m so fond of using.
I’ve been thinking non-stop about a part of that Cory Doctorow post above—subtract the hype bubble, and we’d all be talking about AI’s applications as “plugins.” We’d see it as the normal software that it is. I think this plugin patter has potential to move the bubble a little further towards the needle we’re all waiting on. No-one’s ever thought that a plugin is an independent agent capable of thought and intentional decision-making. Except, of course, for true believers and bagmen.
Keep in mind that the reason we’re in this bubble is not solely because a lot of people believe the bagmen will build god after investors spot them another $100b (and, failing that, Productivity will at least be 1000x)3. True believers are downstream of the actual, material requirements that the bagmen have suckered investors into paying for, hand over fist. Y’know—the hyperscale data centres that require us to xenoform the planet to the point of mass extinction4.
Bosses are revealing their preference for mispriced goods, because, to cite Nikhil Suresh, Spending A Lot of Money transmutes into Something Substantial Being Done:
https://ludic.mataroa.blog/blog/brainwash-an-executive-today
In fact, some of ’em don’t even use the technology that is supposedly “changing everything:”
There have been several occasions where I have seen someone, apropos of nothing, blurt out almost word-for-word “AI is changing everything”, only to concede moments later that their organisation does not currently use LLMs for anything, and indeed, that they cannot name a single thing that has changed other than they get some use out of ChatGPT (frequently the free-tier).
Nikhil Suresh, AI Mania is Eviscerating Global Decision-Making
As for the bagmen, the evangelists, who we do not want to invite to dinner, who poisoned our well with mercury? I think they would really fucking hate it if the rest of us woke up tomorrow, and dug another, unpoisoned well.
So let’s make a start: AI is plugins. Plugins you can run locally on your phone or laptop, powered by your new plug-in solar panel (new if you’re in the UK, anyway5). Plugins that are optional – by definiton, plugins are self-contained: unessential for the main program’s function. Retroarch’s got plugins for filtering audio/video—a nice thing to have, if CRT scanlines on your old games tickles your brain the right way. They don’t tickle mine, so I don’t use them.
Wee, optional plugins, that will stick around after the bubble. Find typos. Transcribe my handwritten notes. Mask this object from the background. Generate a timeline of scenes and then a heatmap showing where, when, and how often a character appears. Find the post where I wrote about XYZ.
Per Charlie’s post:
[A] decision-support tool, not as a substitute for doing the hard work…
The Luddites didn’t exactly wait until child labour was illegal to talk about how steam-powered looms would improve their work. So, I want to ask what the vibes are on “AI is plugins.”
If we accept that handwriting notes, taken USING OUR BRAINS, is a form of research, and that said research can then be digitised with a locally-stored LLM that can perform optical character recognition, then perhaps we could start to be very fucking annoying and subvert common talking points – using AI for research, for example.
As researchers, fiction writers may be tasked with learning about a bunch of niche topics in a short amount of time. A research session may start and end at Wikipedia – or, we may have to descend into the thicket of esoteric knowledge commonly referred to as… “References.” For example, take crop rotation, something I know very little about:
https://en.wikipedia.org/wiki/Crop_rotation#References
As the writer of stories, there are some baseline propositions for research that should be self-evident:
- I am NOT an expert on crop rotation;
- As a writer, I am the expert of my current WIP, and I think that a yet-to-be-learned aspect of crop rotation might affect it; when I find that aspect, I will then decide what information is necessary to show to the reader;
- I am going to be making a simplified model in my head of crop rotation processes, and I provisionally trust the knowledge of human experts who are concerned about the truth – farmers, historians, government documents, nonprofits, etc. etc.;
- The experts’ knowledge may be supplanted in future by better, more refined knowledge;
- The mean time to my worldbuilding being corrected by an expert is well within my lifespan.
An LLM, regardless of its parameter count, has no capacity to care about factual truths. It does not “know” what “truth” is in the same way that we do, and therefore defaults to botshitting6. It is not, in fact, an inexhaustible, always-online PhD holder that knows everything. We do not converse with these things. Fiction writers are experts enough to understand this. What some LLMs can do is find typos and transcribe handwriting/audio:
Again: I am running this model locally, on my low-end Thinkpad. It takes less CPU power than needed to watch a video on YouTube. I do not, under any circumstance, “gotta hand it” to Sam Altman and the Silicon Valley technofascist troupe. Neither do you. They, and their products, are owed a steeped cup of bitter Fuck All.
If an undecided person on the street were to now ask me if I use AI for research, and the terms of their question are as vague as the previous clause (what kind of AI technology? what specific research tasks?), I can get ahead of the conversation, and steer it away from what Emily M. Bender describes as the “untenable middle ground:”
As I see it, I now have a way to persuade that person of what this technology does to me, what it might do for them, and what it is/isn’t. I can show them what parts of the writing process I value, and explain why it is important that I do not shunt these off to the LLM. LLMs that do OCR is plugins. OCR existed before the “gen. AI” well-poisoning, and it will continue to exist long after the bubble bursts.
When you and me make our anti-AI noise in response to the newest scandal, what we’re really doing is standing up for our belief in ourselves; we should get to decide how this optional technology should be used. We love being centaurs7 – we wouldn’t be writing and reading on our PCs if we didn’t! We also love using multiple writing mediums. Said PCs have never replaced our want for notebooks.
Many of us will find AI as optional as a CRT filter. None of us want to be forcibly marched to a hellish reverse-centaur future where we slavishly edit the abysmally dogshit first drafts that PenguinGPT churns out at superhuman speeds. That is deeply inhuman, inhuman all the way down, as mushy and cliché as that is to say.
BUT. (Aha, there it is! *CinemaSins DING DING DING*)
Powerful AI evangelists—the oppressors—deeply detest that cliché. They despise that we think ourselves smart enough to scrutinise their bullshit; that we would dare to decide. We’re not allowed to do that. They say if we don’t talk to god, we will be left behind. They say AI is not just the future—it’s everything. And we should want more of it, forever. 📈🛐
BUT, if we decide that AI is plugins, is optional, and give those idea some airtime… Well, it’s not going to burst the bubble single-handedly – that’s naively hopeful with an extra helping of naivety – but I think the airtime might carry it a little further towards the needle we’re all waiting on8. Take the words out of their mouth, twist them out of context, and use ’em to our own ends. We can do this without sounding like a both-sidesing journalist.
I am, in general, anti-AI. But I do like a good plugin. They’re quite nifty for bringing my handwritten journal and Obsidian vault closer together.
So: should we talk about plugins? Or should I just shut the fuck up?
- Image credits:
mainframe: Hufvudstadsbladet CC-BY-SA 4.0
hal eye: Cryteria, CC-BY 4.0
apartment interior: Yelderberry, CC-BY-SA 4.0
MS data centre: Husky, CC-BY 4.0
↩︎ - Context: my original BlueSky thread, I referred to the user in the Threads screenshot that Charlie reposted. I have no idea who they are, and I ain’t going to Threads to find out. ↩︎
- Do feel free to correct me if I’m wrong – in comparison to the people I’m quoting, my understanding is not so broad. ↩︎
- A.M. must feed the killing data so he can make the Digital Circus. Or something like that. ↩︎
- New by the end of August, according to the Guardian:
https://www.theguardian.com/money/2026/jul/23/uk-households-free-install-plug-in-balcony-solar-panels ↩︎ - Bullshitting, as Harry Frankfurt described it, is also a very human activity. Even if bullshitters don’t care about the truth, they still intend to deceive to their own ends.
Point is, I do not think that an LLM can augment and automate research by guessing what word comes next in a summary. Researchers do, too:
https://cruxevals.com/crux/can-ai-agents-conduct-research/
At best, it might be able to generate a list of references, given a specific question and context. I don’t know for certain. I am satisfied using my eyes, brain, and Wikipedia. ↩︎ - per Cory Doctorow: “experienced workers who are assisted by automation on terms that they set for themselves:”
https://pluralistic.net/2026/08/01/dare-snot/ ↩︎ - For clarification: I do not know if the demand signal for AI can be simply killed by making normies aware that they can install an LLM directly on their phones and laptops. But it might help. ↩︎
- visor/”bucket hat”: Jjanhone, CC-BY-SA 4.0 ↩︎
TTLY… (permalink)
- [Jul 26 ’25] Dan Davies’ “The Unaccountability Machine” https://brologue.net/2025/07/26/440-squirrels/
“Vibechecking ‘AI is plugins'” is licensed under a Creative Commons Attribution 4.0 International license. So long as you clearly link back to this webpage and credit me, bm, you are free to use this post however you like. Quotations, images and screenshots not included; proceed with caution.