This is Contraptions, a weekly newsletter by me, Venkatesh Rao. As the name of the newsletter and the samples below suggest, the general idea is to explore our world as a contraption-of-contraptions, with contraptioney mental models, and a contraptioneer sensibility. The sensibility is that of half-assed engineering, but the scope is universal. I talk about contraptions of all sorts, not just technological ones. So you’ll find discussions of political, economic, and philosophical contraptions.
As of September 5, 2026, this newsletter is fully AI-native. This means AI is used extensively in all stages, from ideation to the writing itself, making the writing itself contraptioney. You should assume by default that everything you read here has some degree of AI involvement. See the AI use philosophy section (really a manifesto-ish mini-essay) below for details.
Sample these posts to get a sense of the themes.
This incarnation of the newsletter is relatively new (I adopted the current name in October 2024), but the newsletter itself started in 2015, and has been through two previous incarnations. More on the history below if you’re interested.
Currently, I publish one article a week. I am not currently paywalling writing, but paid subscriptions are turned on if you’d like to support the newsletter.
I take 1-2 breaks per year, during which billing is paused. For paying subscribers, the next bill date is pushed out appropriately (it’s unpaid vacations from this newsletter basically).
Some History
This is the third incarnation of a newsletter that has been going continuously since 2015.
Between 2019-2024 it used to be called the Ribbonfarm Studio newsletter, and was a kind of Substack expeditionary outpost of my now-retired Ribbonfarm WordPress blog (still online but archived).
Before that, between 2015-19, it used to be called the Breaking Smart newsletter, and associated with another retired project of mine. You can find an ebook compilation of the 2015-19 newsletter issues here.
The idea with this new name and third incarnation was to give myself a 50th birthday present in the form of a curated mostly blank canvas, selectively carrying over a few themes from my older projects, but otherwise trying to engineer contraptioneer a fresh start for myself.
AI Use Philosophy
Between early 2025 and September 5, 2026, I was experimenting heavily with AI, and explicitly declaring AI-use at the essay level, usually sharing prompts and methods, and segregating AI assisted essays in an opt-in section Sloptraptions.
Starting with my September 5th issue, The Curiously Playable Universe, my first “production” AI-assisted post, I am no longer doing these things. There are no article-level declarations or labels, only this philosophy statement.
The technology feels mature enough for production use, and for my kind of writing at least, no longer needs special treatment or expectations-setting that is different in any way. The default has shifted — it is now fully hand-written work that is exceptional. But I don’t flag that either.
Use Declarations
My original motive for declaring AI use, sharing prompts, and segregating AI-assisted posts was to encourage and help others experiment along with me, borrow my techniques, and consciously work on developing an appropriate new literacy in this new medium, for both reading and writing.
This frontier-exploration and literacy-fostering stance was frequently misunderstood as accommodation of prejudices in favor of “purely human” writing, tacit admission of poorer quality (like TEDx vs. TED), and even misguidedly praised as some sort of ethical disclosure practice, which it has never been.
I did not mind being misunderstood this way (or rather, was resigned to it), but to be clear, I do NOT believe a principled ethical stance requires disclosure of AI use, and never did.
As a writer, I stand behind what I write the same way, regardless of how I produced it — shiftily, inconsistently, and unreliably. Labeling was never more than a courtesy to readers, in case they needed a cue to adopt a different reading posture.
It is a courtesy that I judge to be no longer necessary.
The literacies necessary for reading AI-assisted writing have become sufficiently widespread that we can simply go back to thinking in terms of good vs. bad writing as we used to. Declaration of AI use is no more necessary than declarations of Google, library, or spellchecker use.
The primary reason is that as the technology has improved and evolved, the trustworthiness of generated writing has become impedance-matched to my own (if you too are a writer, ymmv). I’ve always been at least as sloppy in my writing as modern AI, and often significantly more so. If you’ve been reading me for any length of time — and I’ve been writing continuously online since 2007 — you know what you’ve been signed up for. Painstakingly overwrought prose, lovingly and reverentially rewritten a thousand times ain’t it.
Our contract hasn’t changed. It simply no longer needs special clauses for AI-assisted things.
At the same time, tools to detect AI writing and filter it out have matured, and are natively supported on Substack, so readers who are still acquiring the necessary new literacies no longer need helpful signage from writers to navigate the emerging world of text. If in doubt, use Pangram and change your behaviors accordingly. Just because it was invented and intended as a witch-hunting tool doesn’t mean you can’t use it as a remedial literacy aid.
Model training protocols have also matured, and are increasingly able to detect generated text and make appropriate accommodations to avoid syndromes like model collapse. This has frankly been worrying me much more than anti-AI reader prejudices — I do not want future versions of these wonderful tools to suffer because my usage is insufficiently tagged.
The only reason to do explicit labeling now is performative “AI use ethics” signaling, (which was never one of my reasons), with the signaling declaring what I consider to be the wrong ethical stance today — one that confusedly and incoherently constructs AI use as something intrinsically shameful, deceitful, and anti-humanist. Something to be subjected to witch-hunting, burning-at-stake attentions, if not penitently declared upfront.
Well, I’ll cop to the third of those charges at least. I’m pretty openly and unapologetically anti-humanist, which is not the same thing as anti-human.
The witch-hunting of course, does not bother me at all. It only works on people trying to “pass off” their work as fully human, to audiences that have decided to care about it.
And certainly, what you can do with AI, especially if you’re entirely uninhibited about it, increasingly looks and feels like real witchcraft. So I don’t blame the scared villagers for clinging to their ineffectual witch-hunting practices. Many are probably going to get hurt, but actual witches and warlocks need not worry.
AI Witchcraft Style Principles
Setting aside the endemic culture wars over AI use, on a more fun note, I’ve learned a lot through my year-and-half of experimentation with Sloptraptions. I’ve mostly satisfied my curiosities now about what’s possible with the technology in its current stage, and gotten comfortable in a new authorial mode.
My most important style principles concerns things I don’t do.
First, I don’t try to make AIs sound like I used to in my pre-AI work. I do try to do that in constructing the voice of the RAG bot trained on my writings, where the conceit is that you’re talking to my digital twin, but it is not an important consideration for new essays. There’s nothing wonderfully unique about my voice that’s worth enshrining in AI-assisted writing protocols. If I have something interesting to say, I’m fine sounding like an index-fund voice.
This means in any given post, you’ll get some idiosyncratic mix of my personal voice and the LLM’s voice (usually the current version of ChatGPT), depending on how I generated it.
Second, I don’t try to edit away or even moderate the native stylistic quirks and tics of LLMs. I largely leave all the well-known markers alone, unless they make it personally hard for me to read the essay. So you’ll find not X, but Y constructs left largely untouched, and LLM-favored word choices like load-bearing, structural, hinge, or metabolic left alone. These quirks evolve with model versions, and I take note of them as they change, but don’t do anything unless I actively disagree with the choices for particular reasons. For example, I force longer paragraphs over the highly “ventilated” prose AI likes to generate, with lots of short sentence fragments and one-line paragraphs since I think it improves readability for close readers, even if it lowers scannability for skimmers.
After all, I don’t try to micromanage human collaborators, so why should AI collaborators be any different?
This particular species of collaborator is in fact valuable precisely because it possesses a voice that is a distilled embodiment of the entire contemporary wild zeitgeist itself. Shaped by training protocol choices, yes, but still, a remarkably distinctive voice that I feel it would be a pity to reshape too much to conform to my more human preferences.
You wouldn’t cook fresh vegetables to death and overwhelm them with aggressive spicing would you? No. You’d try to bring out and preserve their natural character as much as the recipe allows.
To Pay or Not to Pay?
I no longer paywall my posts. This has nothing to do with AI. I just decided, in late 2025, that I didn’t like where paywall culture was taking us. I still like making money from writing. I just don’t like forcing the writing into paywall-shaped designs.
So no paywalls, but paid subscriptions are still on, and you can choose to be a paid subscriber. Doing so obviously supports me in doing more of what I do here, and helps normalize and correctly value AI-native writing.
If you’re neither opposed to AI use as a matter of principle, nor use it as a high-likelihood indicator of low quality and therefore to be pre-emptively filtered out, you have a real choice to make here.
Whether you decide to support my work should probably be a matter of how AI reshapes the content of what I do, and whether it still delivers whatever you’ve come to expect of me, based on past writing.
Here’s the state of play as best as I’ve become aware of it.
Historically, in my pre-centaur era, my writing has been both praised and criticized for 3 attributes, and the question is what AI does to each:
Personal idiosyncrasies, such as gleeful overuse of 2x2s, caricatured archetypes, frequent crossings/recrossings of STEM-humanities divides, a tendency to mix high and low culture, and a great deal of coining of terms.
Generational tendencies attributable to being a GenXer, schooled in our justly-caricatured dispositions — ironic distance, a taste for nihilistic satire, mercenary hypocrisy, and an inability to perform the sorts of overwrought affective sincerity that are increasingly the norm today.
Foolhardy grand ambitions, manifesting as a tendency to make up wild, speculative schemes, theories, and trend hypotheses about the world, based on questionable assumptions, flimsy evidence, and overindexing on things of marginal salience. The sole justification for such indefensibly sloppy proceedings has always been that it led often enough to entertaining food for thought.
I find that writing with AI significantly tempers the personal idiosyncrasies, and dilutes the generational tendencies (the damn Millennials have contaminated the training data with too much sincere authenticity), but significantly amplifies both taste and capacity for grand ambitions.
AI, in other words, better equips me to foolishly rush in where angels fear to tread. But it likely makes my work less personally relatable, and with less of the characteristic texture of fully hand-crafted writing. If those things matter a lot to you, I’m sorry. You’re out of luck. You’re going to get less of that in the future.
For me personally, the first two characteristics, even if valued by readers, were never particularly important. It was all about the third characteristic. This is perhaps why I experience AI, as a writer, to be pure upside. I’ve lost very little of what I care about, and gained a lot of what I do care about.
I can now trespass rudely into domains I have no business getting into, dive deeper into details I can’t properly comprehend, and ascend higher into abstractions too rarefied to breathe in unassisted.
Which means I can do ever more foolhardy things with the same level of personal risk-taking.
Since AI can strengthen and harden elements in my contraptioney edifices that would previously have been weak and soft, I can build bigger and crazier Rube Goldberg structures without adding significantly more fragility.
This means I can, boldly muddle through where unassisted-me could never muddle through before.
If you think the world needs more of this kind of thing, consider becoming a paid subscriber.
