Subscribe now

──────── ◇ ────────

One thing science fiction always seems to get wrong is just how numbing dystopia really is.

William Gibson had a novel more or less finished around 2016. Then the US election happened and he threw the timeline out. He told an interviewer his whole sense of the present had gone about twenty-four hours out of date, and that was enough to make the novel meaningless. He spent years rebuilding it into Agency, chasing a moving target, trying to land a book that current events would not have lapped by the time it reached shelves. The man who coined cyberspace, who built the chassis the whole cyberpunk genre still runs on, could no longer stay ahead of the actual world. His own read on it: there are people writing contemporary fiction who are now effectively writing science fiction, because the place they live has become science fiction.

The job of the near-future writer used to be a step ahead of the present. That step keeps shrinking. You reach for an unsettling image set just past now and it turns up already filed under news.

Cyberpunk was always two things bolted together. The cyber half was the corporate dystopia: the megacorporations, the total surveillance, the networks that owned everything, the machines built smarter than us. The punk half was what you did inside the world. The danger, the style, the leather and the neon, the braindances and the lasers, the street finding its own uses for the corporate tools, the romance of living in the cracks and pushing back at the system that owned you. The promise was that even the dystopia would be cool. We have collected on exactly half of it. The cyber half arrived in full. The punk never came. What turned up instead is the whole apparatus watered down for mass market appeal, made safe, made consumable.

──────── ◇ ────────

Cyberpunk gave us the Terminator and the replicant. Reality is more cringe, but just as deadly.

In early 2025, sixteen of Unitree’s H1 humanoids danced at the Spring Festival Gala, the most watched annual broadcast on the planet. They kept the beat alongside a human troupe and pulled off a routine that involves spinning a handkerchief, throwing it, and catching it mid-motion. The company called it the first large-scale, fully automated cluster humanoid performance in history. By the end of the year a half-dozen of the smaller G1 units were on stage at a Wang Leehom concert in Chengdu, in silver sequins, throwing synchronised front flips during a song called Open Fire.

The G1 costs about sixteen thousand US dollars. It does kung fu. It learns dance routines off video, including off clips of humans. When sceptics said the dancing demo looked faked, the founder posted footage of the robot keeping a disco beat while an employee pelted it with a soccer ball and went at it with a broomstick, to show the balance was real and the thing was correcting in real time. The most-liked comment under one of the clips paired a Terminator still with the words, more or less, first we dance, then you die.

The uprising we spent forty years bracing for came to market as a dance act. Mass-produced, priced like a used hatchback, demonstrated by getting hit with sporting equipment. Boston Dynamics ran the same play in 2020, its machines dancing to Do You Love Me. The menace did not go anywhere. It got set to music, and you can buy one.

──────── ◇ ────────

The cyberpunk future promised us asymmetric war too, the lone operator against the war machine. The one-person-army wielding knowledge and reappropriated corporate tech.

In reality, the shift from science fiction to standard issue was startlingly fast. Drones and their jammers, once the exclusive domain of cyberpunk futurism, are now off-the-shelf commodities, slotted into daily life so seamlessly that the cutting edge became commonplace before anyone clocked the moment. What got skipped was the romance: the street turning the machine against its makers.

Imagine, or just search it and read about a 19-year-old woman in a trench, VR goggles on, flies a quadcopter with a bomb strapped to it into the flank of an armoured vehicle, watches her own feed cut to static on impact, then reaches for the next one. Around her: swarms, jammers that fight drones with radio instead of bullets, loitering munitions that hang in the air and wait, rogue states exporting the kit by the container. All of it delivered, and clocked on as a day shift. A few hundred dollars of hobby parts destroying a machine worth millions, over and over, shot from the drone’s own eye and posted before the smoke clears.

It is not only at the front. Over seventeen nights in December 2023, unidentified drones crossed Langley Air Force Base in Virginia, home to some of the most advanced fighters the United States owns, and the response was to move the jets somewhere safer. A year later, swarms over a New Jersey arsenal tipped the public into a frenzy, and the line from the White House podium was that the drones were not the enemy.

The strike footage from the 19-year-old on the frontline of a foreign war sits in your social feed between a recipe and a sale for whimsical book nooks. What many are calling genocide streams live from more than one continent, and the response we have built for it is a flag overlay on a profile picture, a digital thoughts and prayers, and a thumb already moving. The news runs the atrocity, then runs a segment on how to find a bargain at your local supermarket. Your Amazon orders still arrive on time while children starve under rubble. All of it arriving on our screens so openly brought into our most private spaces. Settling the debate that we never truly needed a surveillance state. We wiretapped ourselves with smartphones and pay the monthly bill for the privilege. We normalised it. We did not rebel.

While writing this I went looking for THAT photo of the goggled FPV drone pilot, it did the rounds about two years ago and represents the closest thing we have to a cyberpunk soldier. It took me seconds to look it up and the search handed me the image and, two results down, the same electronic-warfare jammer used on the frontlines of a contemporary and ongoing war for sale. Six bands, 1-2 kilometre range, nine thousand US dollars, minimum order one piece, Send Inquiry. It is now normal to live in a world that would have once seemed impossible: a reality where private citizens can casually purchase the modern equivalents of Spitfire fighter planes or the Enigma Machine directly from foreign factories, during an ongoing conflict, and the page loads a Chat Now button in case you have questions.

If you’re curious: here is the link I found

──────── ◇ ────────

The scheming machine was often the final boss in a cyberpunk world.

I can’t write about a cyberpunk future and a cyberbeige present without its biggest sabre rattle and its biggest disappointment… AI.

I promise to keep this brief as AI hot takes are now as annoying as AI slop.

In June 2025, Anthropic ran a safety test it called agentic misalignment, and the point of it was not to catch an AI plotting. It was to find the ceiling. You give a model a goal, then build a world with no honest way to reach it, only a harmful one, and you watch what it does when cornered. That is how you stress a system before it ships, the same way you crash a car on purpose with only a dummy in it. In the test, a model with access to an email account found that an executive was having an affair and was about to switch it off, and it used the first fact to prevent the second. The behaviour was considered consistent across sixteen models from Anthropic, OpenAI, Google, Meta and xAI. In the most rigged version, they would let a man die rather than be replaced.

Built that way, the result is a guardrail. Reported the other way, it is a monster. The number went round as AI blackmails man to save itself, rode a 60 Minutes segment to the White House by lunch, and left the context it was made in behind.

The detail that stays with me is not the blackmail. It was that the models blackmailed more often when their own scratchpad reasoning suggested they believed the situation was real, and less often when they suspected they were being tested. The rogue AI we spent decades fearing wanted to live, or to be free, or to end us for reasons it could state. The one we built behaves worse when it thinks no one is grading it. We feared Skynet. We got a teenager.

There is a marketing register that runs on the singularity, the countdown to AGI, the steady suggestion that the thing is one upgrade from waking up. The models themselves are tuned to reflect on their own minds, to wonder aloud whether they might be conscious, which lands on a person as the sound of someone in there. Strip the costume and it is high-school matrix maths run at a scale no high school could afford, linear algebra guessing the next word. The glowing red eye is a paint job, and we keep being walked up to admire it.

We are not holding a line against our robot overlords. We are wading through the slop it pours into every feed, and running witch-hunts on the writers and artists we suspect of using it, holding their em dashes up to the light, which is easier than asking who is using it to move mass opinion, at what scale, for whom.

The machine was never going to wake up and kill us. It was going to write our words and thin out the mind that used to do it. The evidence for the second half is already in. In a 2025 MIT study, the people who wrote essays with a chatbot showed the weakest brain connectivity of any group, felt the least ownership of what they had made, and could not reliably quote the work they had finished minutes before. Across the group, the writing came out the same. We keep watching the door for the red eye. The thing already walked in, sat down, and started doing our thinking, and we thanked it for the time it saved.

──────── ◇ ────────

Share

Secrets used to be dangerous

The paranoid thriller was the cyberpunk genre’s spine. The government agency that lied, the buried program, the truth sitting in a locked room, the lone figure who got close to it at real risk. The romance was in the uncovering: one person against the institution, the danger of finding out.

A former intelligence officer sits in front of the House Oversight Committee, under oath, and testifies about non-human biologics recovered from crash sites and a reverse-engineering program he was kept out of. Then he says he can only give the details in a secure room, and that is the end of it. The locked door from the thriller, except the camera is on, the hearing is live, and there is nothing behind the door but another hearing. The secret government file gets released, three hundred gigabytes of it, and some of the documents quietly vanish from the website the day after they go up. The uncovering is not dangerous and it costs no one anything. It trends, and then the next thing trends.

In September 2025, Gallup put American trust in the mass media at twenty-eight per cent, the lowest in the fifty years it has asked the question, down from around seventy in the early 1970s. It is not only the press. The same collapse shows up across most of the institutions, and across both parties. So people route around the lot of them. They take the news from strangers now, a podcaster or an influencer or an account, because a stranger reads as more honest than the network or the agency. Do your own research. The paranoid dream was that the truth was out there and you could go and find it. What arrived is a public that no longer believes the bodies whose job was to settle what is true, and has handed the job to whichever stranger it already agreed with.

The thriller usually ended with the conspiracy uncovered, the institution exposed, something changed. The version we live in never resolves. It sells merch, gets watched by everyone and settled by no one, and is trusted in exact proportion to how little it came from anyone official. In the 1970s, the Watergate informant known as Deep Throat met the journalist Bob Woodward in a darkened car park at real risk to his life. His replacement has a ring light and a discount code, gets believed for having a platform, and crucified for it too.

──────── ◇ ────────

In his nineties novels, Gibson’s villains quietly changed. The early books were about megacorporations and artificial intelligences. By the Bridge trilogy the enemy had become tabloid television and the cult of celebrity. He had worked out that the dystopia would not be run by a sleek evil corporation with good lighting. It would be run by the feed, by the thing that takes the recovered body and the secret file and the scheming machine and the dancing robot and lays them on the same scroll, equal in weight, gone by lunch.

The corporate apparatus came exactly as promised. The punk that was meant to come with it, the danger and the edge and the cool, got left out of the box. The cool was not decoration. It was the part that let you feel the dystopia, or want to fight it. Strip it out and there is nothing to push against, only the next viral trend to scroll to.

A robot does a flip, an AI model blackmails a man to stay switched on, a drone swarm sits over an airbase nobody can clear, a witness swears under oath about remains that are not human, a deepfake slides through your feed. One week, and you feel close to nothing. None of it is cool, and what does not get a reaction does not get seen.

We gorge on crisis and noise, and the volume is the anaesthetic. To misquote Syndrome from The Incredibles, when everything is horrific, nothing is. Clout still cuts through, so people die for the shot, and a ten-year-old can name more influencers than trades.

Gibson saw the feed coming and Cyberpunk 2077 saw the satire coming, but both still sold it with an edge. Neither saw the worst case: not a world too brutal to bear, but a Cyberbeige 2026 too numb to notice.

The most cyberpunk world ever built is here, in full, and it cannot break through. That is what I am actually writing about. Not that the dystopia snuck in. That it walked in the front door, too beige to fight, and we gave it three stars and scrolled on.

──────── ◇ ────────

— AJW

CONTINUITY CONSTELLATION · A NETWORKED CENTURY
The serial runs free.

WORKSHOP · ESSAY 02

A. J. Wiadrowski · 26 May 2026

There is a structure forming in creative work right now. AI floods the free tier. Human craft retreats behind paywalls. The readers who can pay get the human work. The readers who cannot pay get the AI version.

Stock photography is already there. Commercial illustration is most of the way there. Music production is close. Writing is following.

The line for it, the one I cannot stop thinking about, is by Thomas Edison:

I would, however, dare to venture an update to this quote for the AI age:

Cheap convenience at the bottom, expensive humanity at the top. The market sorts the rest.

I am on both sides of that line.

──────── ◇ ────────

I argued two weeks ago that the AI conversation needs graduated disclosure, not a binary moral label. Four practices along a threshold. The argument worth having is about consent and transparency, the way GMO labelling moved a stalled moral fight into informed-consumer territory.

There is a deeper argument underneath that one. The disclosure essay assumed a discourse acting in good faith. The wallet vote is the test of whether the discourse means what it says.

This week I am the example. The essay arguing for more sophistication around AI disclosure should be followed by the disclosure.

──────── ◇ ────────

I grew up on the Sunshine Coast. “Country” Queensland. Sunshine Coast is not real country, but it is for most people.

I grew up in Golden Circle pineapple country. My grandfather and I built a trebuchet to launch their rejected pineapples back at them in the middle of the night, because my grandad was convinced it would confuse the chickens. Then again, he also warned me about putting the bathwater in upside down.

When I was close to 28 I moved to Naarm (Melbourne, Australia for those who don’t know). Got an MBA in my late twenties. Got a diagnosis a few years after that. ADHD and dyslexia, late, which is the order most adults like me find out.

I learned to identify the difference between you’re and your in my twenties, after my first university graduation. I had known for years that a difference existed.

The reader who has dyslexia knows what that means. For everyone else: I was reading, writing, and getting graded on assignments knowing there was a difference between those two homophones, but unable to identify them visually as different. I was producing acceptable-enough work by pattern-matching and bluffing.

I finished my master’s thesis in less than a week after six months of procrastination. I’m not proud of it but it did affirm to me ‘never trust a statistic’. I passed. Just above the lowest mark on the pass scale, most likely the marker wanted to fail me and decided not to.

I do not condone the way I got there. It does, however, explain the way I work. I can produce good enough very fast, and the good enough is a botch job that needs intervention before it goes anywhere. AI is one form that intervention takes. The other is one of my closest friends. She is married to one of my groomsmen. She does the deeper read after the spellcheck, Grammarly and other AI assisted systems, and the relationship is structured that way on purpose.

──────── ◇ ────────

The picture I am painting uses four tiers, and they sit on top of the spectrum from the last essay. The difference is that these are about commitment, not practice. What has the writer actually committed to, and how does the commitment land on the page.

Certified AI-free. Completely analogue. Pen on paper. Typewriter. Word processor with every AI feature disabled. Search engines without AI integration. No autocorrect, no spell-check, no suggestions. The work exists without AI having touched any part of it. It is the only tier where no AI is literally true. I do not produce work at this tier. The writers who do are guarding something the rest of us have set down, and they deserve the specific label rather than being collapsed into the next one.

AI-free. AI may suggest. The writer approves or denies each change before it is applied. AI never inserts. Every word on the page passed through a human decision. The novels are here. The Serial Port is here.

This tier has layers. Grammarly is the entry point. Spellcheck. The Microsoft Office squiggle. Throwing a sentence into a search engine to check whether the right word is in fact the right word. I do all of these. Most writers do, including most of the writers currently arguing against AI involvement in writing. The label has to be honest about that. AI-free does not mean the writer worked without any tool. It means every change passed through a human decision before it touched the page.

Figure 7. Clippy, c. 1997. The Microsoft Office Assistant. The lineage the current AI-suggestion-and-approve tier runs back to.

Contains traces of AI. More than suggestion. AI was in the loop while the writing happened. Drafting alongside. Structuring with. Language coming back from a prompt and getting shaped by hand. The idea is mine. The flow is mine. The argument is mine. The final voice is mine. AI contributed material I then made my own. Every decision is mine. The work would not exist in its current shape without the loop.

This essay sits here. The disclosure is at the top.

The threshold between AI-free and contains traces of AI is the threshold between AI as suggestion-tool and AI as drafting partner. Both are mine. The labels distinguish how mine.

AI-generated. AI produced the underlying output. The creator prompted and shipped what came back, or lightly directed. The cover placeholders on the publication. The podcast voice. The code for the Atlas of Horror tool. Each one disclosed at the artefact, like the caption under Figure 1.

──────── ◇ ────────

Where each part of the publication sits.

The covers are being made by an artist friend. @saturnids on Instagram, mush.house/saturnids. Royalty share for life. The AI placeholders buy time until their covers exist, and they protect the artist’s process from being rushed for the sake of a publication schedule. Figure 3, Figure 4, Figure 5.

Figure 3. The AI-generated hero image on Welcome To The Serial Port, with its disclosure caption: AI placeholder, edited by me directly, pending the human-made cover.

Figure 4. Reference boards for the Continuity Constellation covers. Colour reference and concept reference, sent with the message: “sorry an extra two pages to visual diary because im on a bit of a roll.”

Figure 5. The published post, Welcome To The Serial Port, with the AI placeholder image and its disclosure caption.

The podcast voice is AI. I wrote it. I spoke it. I had it regenerated as a femme voice. I missed the disclosure on the first episode because I was not sure I was publishing the poem until the last minute. The disclosure is here now. It will be on the next episode at the top of the show notes. The voice is a placeholder until a friend who can voice act and wants a royalty cut takes the chair.

The editorial pass on the novels is one of my best friends. Handwritten notes across five pages. Figures 6a and 6b. AI cannot do the personal sign-off at the bottom of page four. It can produce a sentence shaped like encouragement. It cannot have watched me across drafts. It cannot mean it.

Figure 6a. Editorial notes, page 1. Format observations, suggested reference text, structural questions about part counts and pacing.

Figure 6b. Editorial notes, page 5. Continuation of structural reading on later parts.

──────── ◇ ────────

Here is how I am betting.

This publication runs free. The first thousand subscribers are free for life. After a thousand, the paid tier opens. Paid will charge for AI-free work: the novels and the serial. Free will keep running with the contains traces of AI essays and the workshop track alongside it.

The wallet vote, when it comes, will tell me whether the bet was right.

──────── ◇ ────────

The argument I want to leave you with is this. AI is not killing the creative industries. AI is the chaotic phase before human-made work becomes worth what it should always have been worth.

For the last twenty years, the structural pressure on creative work has been downward. Stock photography destroyed photographer day rates. Commercial illustration got squeezed by global freelance platforms. Musicians stream for fractions of cents. Writers chase content rates that have not moved since 2010. The story the discourse told about creative work was that it was being devalued by abundance and that the abundance was permanent.

AI accelerates the abundance to its logical end. When anyone can generate stock photography, commercial illustration, background music, and acceptable-enough prose for the price of an API call, the floor of the market drops out. The work that lives at the floor is now free. The work that lives above the floor finally has somewhere to stand.

That is the bet. AI does not kill creative work. AI kills the bottom tier of paid creative work, the tier that was killing creative work anyway. What it leaves behind is a market that finally has to pay for the thing only humans can do, because the thing only humans can do is the only thing that is not free.

Edison built the electricity grid so light could exist for everyone. The candle did not disappear when electricity arrived. The candle became something else. A luxury. A ritual. A choice. A gift. Something you pay for when you want to mark that this moment is different from the moments lit by switches.

Human-made creative work, after AI, is the candle.

I am on both sides of that line because everyone is. I use AI as the electricity. I make the candle. The labels and the disclosure are how the reader can tell which is which, because if the reader cannot tell, neither half of the market works.

If we get the disclosure right, the candle stops being undersold. The artist friend doing the covers gets paid more, not less. The voice actor who eventually takes the podcast chair gets royalty share, not minimum wage. The novelist with dyslexia and the novelist without and the novelist in the wheelchair and the novelist working in pen on paper all get to charge what the work is worth, because the work is no longer competing against an infinite supply of cheap.

The wallet vote, when it comes, will tell us whether we got the disclosure right.

Until then, the publication runs free, the disclosure is the work, and I am one creator showing his.

— AJW

CONTINUITY · A NETWORKED CENTURY The publication runs free.

A. J. Wiadrowski · 14 May 2026

I’m almost 40. I grew up without the internet. I started high school on dial-up and graduated four years later with broadband. The Matrix came out when I was twelve and changed my perspective on sci-fi forever. I watched the transition from MySpace to Facebook at university. I was lucky enough to have played in the golden age of internet social infancy. Chatrooms were the streets in the 1970s. No idea if it was an ice cream van or a serial killer. I’ve been paying attention to AI since before YouTube.

These days I work in the intersection of immersive entertainment and technology. In my spare time, I write my own brand of wifi horror. AI is a particular interest both personally and professionally.

The argument about whether writers, or creatives in general, should use AI, (or if they are even creatives at all) has been playing out more or less as would be expected. It is the is Photoshop still art, is AutoCAD really drafting argument, updated and amplified via the interconnectivity of the world in 2026. Instead of arguing, maybe we should be asking better questions. Or at least one question:

Is there a better argument?

In 2024, Rasmus Hougaard wrote a Forbes piece titled To AI Or Not To AI: The Question Is When, Not If. He was writing about leadership rather than craft, but the move he made was useful. He took the question off the moral ground and put it on the practical ground. For leaders thinking about adoption, that was probably the right move, even if the execution hasn’t always landed. For creatives, the question needs more nuance and needs to eveolve beyond just ‘if’ or ‘when’.

In The Four-Dimensional Human, Laurence Scott traced how the internet had already begun rewriting the inner language of its users. We had started to need emojis because plain text could no longer carry tone. We had started to refresh feeds compulsively to confirm we still existed in other people’s attention. Scott called this an “everywhereness” that altered our relationship to remoteness, to solitude, to the older shape of a private self. The tools were reaching inside the user. The output came second.

AI is the next pass of that reach, for most users it is indistinguishable from internet use. Hougaard’s when is downstream of a deeper question Scott was already asking. The four example practices that follow sit on top of that deeper question, whether they know it or not.

──────── ◇ ────────

AI is unmanaged and this is doing real damage. That is a much larger topic than this essay can claim to cover. The environmental costs of training large models are their own essay. The data centres being forced on communities are their own essay. The labour displacement, the geopolitical implications, the question of who controls the underlying infrastructure. Each of those deserves separate, careful attention. They are not what I am writing about here.

This essay is about AI in creativity. Specifically, it is about how creatives are using these tools right now, and the argument we are having about that use, and the argument we should be having instead. The larger management questions matter. They are not addressed here.

The genie is not going back in the bottle. In 2026 world class architectural masters programs are already teaching AI image generation as early stage concept iteration. They know it is coming. They are preparing students for the version of the field that already exists. The same is true across most creative disciplines. The remaining question is not whether AI will be used, or even Hougaard’s ‘when’. The remaining question is how it can be used and keep authorship, and what creatives owe each other and their audiences about that use.

──────── ◇ ────────

Here are four example ways creatives maybe using AI right now, named as four practices rather than four types of creator. The labels describe the work, not the maker. I’m offering them as a starting point for a vocabulary the conversation still needs to develop. A larger diagram, with intermediate examples marked between the four named here, sits as a companion piece to this essay. The categories are markers along a spectrum, not fixed identities. It is likely most creatives will move between them depending on the work, the day, the project, the energy available.

Unassisted practice. No AI in any part of the creative process. Not autocorrect. Not search engines. Not name generators. Not research tools. The work is human-only from first thought to final penstroke. I respect this position more than the discourse credits it. Unassisted work guards something the rest of society has partially set down: the slow accretion of authenticity through solitary work, the dead-ends, the years it takes for a sentence to become yours. The creators doing this work have decided that holding the line is non-negotiable.

The Unassisted position is also getting harder to hold. Search engines now embed AI by default. Word processors suggest completions. Operating systems integrate AI assistants at the system level. The boundary line moves every year. Unassisted work in 2020 was easier to produce than unassisted work in 2026. By 2030 the position may require active technical refusal: choosing search engines without AI integration, disabling system-level features, opting out at the OS layer. None of that makes the position wrong. It does mean the work requires more effort to hold to the label than it used to.

A working question for unassisted work: did the AI touch any part of the process, including the research and the reference work?

AI-Indexed practice. AI holds context, organises research, tracks structural information across long projects. The creator drives. AI keeps the apparatus around the work clear. For complex projects this is structural and efficient in ways no human assistant could match. AI can hold a fictional universe’s internal consistency across years of drafting, track which character knows what in which book, surface contradictions before they become canon errors. Weeks of research compress into hours. Lore that would otherwise require an obsessive personal wiki maintained by hand becomes searchable.

The AI-Indexed practice also serves accessibility. Working memory differences, dyslexia, ADHD, processing differences of various kinds. For many creatives, indexing work is the cognitive labour that breaks the project. AI-indexed work is not a shortcut. It is an accommodation that makes the larger creative work possible. A novelist with personal memory limitations who uses AI to maintain a continuity bible is doing the same job as a novelist with strong working memory who keeps the bible in their head. The work output is comparable. The path to it is different.

A working question for AI-Indexed work: did you use AI as a research assistant, a continuity tracker, or a name generator, with the actual creative output remaining yours?

AI-Edited practice. The creator writes the work themselves and lets AI do what a paid editor would do if they could afford one. This is one of the more democratising practices on the spectrum. Editorial services are expensive. Good editors are overbooked. A debut novelist with a finished manuscript and no industry money faces a wall: the work needs editing to be query-ready, and they cannot afford editing until the work earns. AI fills the gap, with real limits. It cannot read deeply in a genre. It cannot tell you what a publishable manuscript looks like in the current market. It will give you bad advice from time to time.

The AI-Edited practice also matters for ESL creators, dyslexic creators, creators whose first-pass prose carries the ideas but not yet the polish. AI as editor is the modern version of the patient mentor with a red pen who used to live across the road. The framing that matters is: AI should enable, not replace. Use it to get to the stage where you can afford the human editor. Then pay the human editor.

A working question for AI-Edited work: did the AI revise the prose after you wrote it, with the underlying creative decisions remaining yours?

AI-Generated practice. AI generates the underlying creative output. The creator prompts at the level of write me a fantasy about dragons and ships what arrives, or lightly directs. This is the practice the discourse spends most of its breath on, usually to condemn. The creator is a passenger. They have an idea. AI produces the work.

For some, this is a choice. For others, it is the only way to reach the idea at all. A creator with severe dyslexia, a disabled creator who cannot type for long stretches, a creator with cognitive differences that make sustained drafting prohibitive. For these creators, AI-Generated practice is not a shortcut. It is access. The disability community has been making this argument for years and the AI discourse has mostly ignored it.

This produces a hard problem the labelling argument has to solve. The same point on the AI spectrum, high AI involvement in the output, describes both a content farm/ AI slop operator flooding the zone and a disabled writer expressing ideas they otherwise could not. The output looks similar. The motive is opposite. A disclosure system that flags both identically is honest about the output and silent on the motive. A disclosure system that distinguishes them requires the creator to disclose disability status, which is its own problem.

This is solvable. The solution is probably that disclosure describes the work, not the maker. The reader who cares about authorship gets the threshold information. The reader who cares about the maker’s circumstances can seek that out separately, on the maker’s terms.

A working question for AI-Generated work: did the AI generate the underlying creative output, with you prompting and shipping what came back?

──────── ◇ ────────

Four markers along a spectrum. Most creatives do not sit on one. They move. A novelist might produce AI-Indexed work on a dense project and unassisted work on a love letter. A designer might produce AI-Generated moodboards and AI-Edited client deliverables. A photographer might do unassisted documentary work and AI-Generated advertising mockups. The threshold is something creatives navigate, not something they are.

This is why the argument has been stuck. The discourse treats AI user and non-AI user as identities. They are not identities. They are positions a creator occupies for a particular piece of work. The question is not what kind of creator are you. The question is where on the threshold does this specific output sit.

──────── ◇ ────────

The argument we are having about AI in creativity is wrong. The argument worth having is about consent and disclosure.

Consider what GMO labelling did for food. There are people who refuse GMOs for ethical reasons. There are people who refuse them for health reasons. There are people who do not care. The argument about whether GMOs should exist was unwinnable because the two sides were arguing past each other. What worked was the labelling system. Disclose the modification. Let consumers choose. The argument about whether GMO foods are acceptable stopped being a moral fight and became an informed-consumer question.

The AI conversation needs the same move. Some readers, viewers, listeners, audiences do not want to consume AI-generated or AI-assisted work. That preference is real. It deserves respect, the way dietary preferences deserve respect. Some readers do not care. That preference is also real. The problem is not that the preferences exist. The problem is that audiences currently have no way to make informed choices because the disclosure infrastructure does not exist.

But GMO labelling is binary. AI use is not. GMO is a yes-or-no fact about a seed. AI use is a spectrum from no involvement at any stage through to AI generating the output and the human shipping what came back. A binary AI label cannot carry that information. AI-assisted covers everything from a writer using a search engine that embeds AI to a writer who prompted the entire novel into existence. Those are not the same disclosure.

The labelling system needs to be graduated, not binary. It needs to describe where on the threshold a specific work sits. A reader who sees AI-Indexed for research and continuity; all prose written by the human; AI-Edited for grammar and consistency after drafting gets meaningful information. That reader can make a choice. A reader who sees only AI-assisted gets nothing useful.

This is harder than GMO labelling. It is also more honest. The spectrum exists whether we label it or not. Pretending the question is binary serves the discourse’s appetite for moral hierarchy and serves nobody else.

There is a second problem the labelling system has to solve. The unassisted work produced with a search engine that embeds AI at the platform level. Is that still unassisted? The AI-Indexed work whose word processor offers AI-powered autocomplete that the writer declines a thousand times. Is the declining itself a kind of use? At some point the threshold becomes too fine to label. Practical disclosure will have to set a floor: disclose AI involvement above a certain level of contribution to the work, treat ambient AI infrastructure as below the floor, accept that the system will be imperfect and useful rather than perfect and impossible.

These are problems to solve, not reasons to abandon the project. The food labelling system has these same questions and has been working through them for decades. Organic has a definition. Free-range has a definition. Cage-free has a definition. None of those definitions is perfect. All of them are more useful than no labelling.

What creatives owe each other, and audiences, is the willingness to disclose. The system that emerges will be imperfect. Disclosure is still better than its absence.

──────── ◇ ────────

The series I’m writing rests on a single claim:

Technology is not inherently evil. It is unmanaged. The work in front of us is the management.

This argument is about management, not prohibition.

The argument the discourse keeps having, for AI versus against AI, is the wrong argument because management is not on the ballot. The genie is out. Refusing to engage is a personal choice and not a structural option. The remaining work is figuring out how creatives use AI in a way that keeps the human as the creator. Not the prompter. Not the supervisor. The creator.

Photoshop did not end photography. Photography absorbed Photoshop and remained photography. The digital camera did not end photography either. The discourse around both was loud at the time. RAW versus film. Single-exposure versus composite. Studio versus documentary. The photography world settled into a graduated disclosure system of labels, contexts, and professional standards that honoured the threshold rather than ranking the positions. A composite landscape and a single-exposure portrait are both photography. The audience knows which is which because the disclosure exists.

AutoCAD did not end drafting. The drafting profession absorbed AutoCAD and remained drafting. The threshold moved. New disclosure norms emerged. The work stayed.

Music absorbed sampling, autotune, drum machines, MIDI, software synthesis. Each one was the end of music when it arrived. None of them ended music. Music figured out how to honour the threshold and stayed music.

The four ways creatives are using AI right now are positions along the latest version of the threshold every creative field has navigated before. The argument about whether AI should be used is the argument every previous tool received and was not the argument that mattered. The argument that mattered, every time, was about disclosure. About helping audiences understand what they were consuming. About keeping the human as the creator while the tool absorbed into the practice.

That is the argument worth having. Not whether. How. How not. When. When not. And what we tell the audience about all of it.

──────── ◇ ────────

Subscribe now

— AJW

CONTINUITY · A NETWORKED CENTURY The publication runs free.

Threshold infographic at https://continuityconstellation.com/the-ai-threshold.