The Minimum Viable Author

I hear work dismissed as “AI slop” all the time, but nobody seems especially interested in defining what that means.

Everyone I know is using AI somewhere in their work. But the moment someone admits that AI helped them write, people treat the admission as proof that the work is fraudulent. Apparently, everyone else is sitting at a typewriter over the weekend painstakingly composing their B2B LinkedIn posts.

Ever embracing the courage to be disliked, I have been unusually open about how much I use AI, in part because I think the secrecy is making this conversation stupider. We need better authorial standards for AI-assisted work and a meaningful definition of AI slop, but right now, we seem to be conflating all of it with the simple fact that AI was involved.

Let's start with the clean case: AI wrote nothing

A conversation with Aaron Black

Last week, I spent an hour talking with Aaron Black, a senior editor at O'Reilly and an avid writer, who has been authoring fiction for more than twenty-five years. Disclosure first, our conversation was about his personal creative work and process; he was not speaking on behalf of O'Reilly. Like myself, he has been unusually open about his process online, which has been refreshing to see. It is genuinely brave that he is willing to explain his process publicly at a moment when admitting that AI touched your work can feel like inviting a witch trial.

When the subject turned to slop, Aaron questioned whether we need the AI modifier at all. Slop existed long before generative AI: work that was rushed, barely reviewed, and produced without much thought.

This is why Aaron is such an unusually good person to have this conversation with. He is an experienced, award-winning author, and even when he mentioned that he used AI in his process, some people still raised their eyebrows and criticized him without really understanding what he was doing.

His own process is about as conservative as AI-assisted writing gets. AI supported his research, planning, continuity, and editing, but it never composed the prose. He wrote four novels in ten months that way.

The witch-trial comparison is unusually literal here. Aaron’s series focuses on Cunning’s Hollow, a fictional Connecticut town whose history reaches back to the seventeenth century.

Four novels in ten months

Cunning’s Hollow is a supernatural series set in an alternate-history Connecticut town. Aaron lives in Indiana and wanted the place to feel close to somewhere he already knew, with the rolling hills and weather of Brown County. But Indiana did not have the older history the plot required. Moving that familiar landscape to New England gave him a place he could still picture clearly while connecting it to the seventeenth century.

Aaron came up with the town, its characters, and the shape. Then he described the environment he had in mind and used AI to identify plausible places in Connecticut with similar terrain, weather, and proximity to a major university.

He also generated visual references for his characters and locations, then reworked them heavily in Photoshop when the details were wrong. He created a dark, wood-paneled academy library and later remodeled the same fictional space into the Cunning’s Hollow Arts Center. A group portrait of the first book's cast helped him keep their ages, appearances, and relationships straight.

He kept the images open while he wrote, so if a character crossed the library, he could see where the desk and staircases were. When he changed the building, he could preserve enough of the original architecture for it to feel like a renovation of the same place.

He built a similar system for continuity. After finishing a chapter, Aaron used AI to extract what had happened into a short reference sheet. He could search those sheets to find where a character last appeared, what happened to an important object, or which chapter contained a particular interaction. When each manuscript is roughly 90,000 words, that saves a lot of rereading.

He does not think the result is simply faster. People who have read the books have told him they are his best work, and he thinks the stronger structure, spatial detail, and continuity are a large part of why. They also recognize them as Aaron's books because the weirdness is still his.

He also experimented with using AI to decide what should happen next.

It was terrible at it.

AI kept suggesting choices Aaron would never make, so he stopped asking it what should happen next. He made every story decision himself, used AI only to organize his rough notes into beat sheets, and updated the continuity record after each chapter.

What matters to me is that Aaron was deliberate. He tested where AI helped, recognized where it made the work worse, and changed his process accordingly. He kept the parts that helped him research, organize, and remember his decisions. He threw out the parts that interfered.

We fetishize the craft of production

I have an art degree, so I know.

For most of history, having an idea and being able to execute it were inseparable. A writer had to construct the prose. A programmer had to write the code. An illustrator had to render the image. We learned to associate mastery of the production process with legitimacy.

AI is breaking that apart.

Someone can have something worth saying without being particularly good at writing. Someone can understand exactly what software should do without knowing how to code it. Someone can have a visual imagination without knowing how to draw.

AI is democratizing execution. And that makes people uncomfortable.

A lot of us spent decades getting really good at skills that are suddenly becoming easier to access. I understand that discomfort. I feel some of it too.

But difficulty is not the same thing as value. Suffering through a more arduous process does not automatically make the result more authentic.

Aaron compared it to buying a chair. Sometimes you want something handcrafted in Amish country. Sometimes IKEA is exactly what you need. And then there is the Temu chair that may or may not hold a person (closer to the ‘slop’ end of the scale).

The point is that not everything needs to be handmade. It does need to do what it claims to do, and somebody needs to have checked.

Successful authors have always had access to help: researchers, assistants, editors, illustrators, and continuity experts that make their worlds concrete. Independent writers historically have not had these luxuries. AI gives individual creators access to some of that support, which to me is exciting and empowering. That does not make everything people produce good or make everyone who uses AI an author.

Aaron drew a very strict line for his Cunning’s Hollow series. AI helped with almost everything around the writing, but did not write any of the prose. On this spectrum, his authorship is unusually easy to defend.

But I personally do not draw the line there.

I think someone can use AI to help write prose and still be the author. But I do think authorial authenticity depends on what they are making, why they are making it, what role the prose plays, and what the audience reasonably believes the author contributed.

I do not expect the same thing from a novel, a Python book, and a LinkedIn post. In a novel, the language itself is a large part of what I came for. In a technical book, I care more that the author knows the subject and can teach it. On LinkedIn, I mostly want the person's real experience or idea. Treating those as the same kind of authorship is where the conversation falls apart.

The standard cannot simply be who assembled each sentence. It has to account for who did the intellectual and creative work.

Which brings me to the question I think we should actually be asking:

What is the minimum viable author?

The five things an author still has to do

There are limits. If I know nothing about Python, ask AI to write me a book about Python, barely understand it, and put my name on the cover, I am not meaningfully the author.

But if I understand the work, develop the ideas, direct where it goes, exercise judgment over what stays, and take responsibility for the result, why should authorship depend on whether I personally assembled every sentence? Indeed, this is the essence of creative direction, and the standard workflow for seasoned professionals with expertise and teams supporting them.

As we try to draw or redraw the line, for me, authorship still requires five things.

1. Authority. You need enough command of the subject or creative world to explain and defend the work. You do not need formal credentials, but the work cannot outrun your understanding of it.

2. Contribution. You need to supply a substantial share of the central ideas. I do not think we can turn this into a clean percentage because ideas are not interchangeable units. One defining insight can matter more than twenty smaller ones. The useful question is whether the work's central thinking would exist without you.

3. Originality. You need to make connections, interpretations, or choices that were not simply handed to you by the system. Originality rarely means creating every ingredient from nothing. It shows up in what you notice, combine, reject, and make meaningful.

4. Deliberateness. You need to direct the process, including where AI belongs and where it does not. You decide what the system is allowed to do, reject what does not belong, revise what is almost right, and determine when the work is finished. You have to actually read and review the work instead of rolling the dice until the system produces something plausible.

5. Responsibility. You need to stand behind the result. If the work is wrong, incoherent, misleading, or harmful, the tool does not absorb the consequences. Your name is on it, so you have to be able to answer for it.

An experienced Python educator could design a learning sequence, supply the examples, use AI to help draft straightforward explanations, test every exercise, correct the mistakes, and accept responsibility for the finished book. The exact sentences might not all be theirs, but the expertise, pedagogy, judgment, and responsibility would be.

That person and the person who knows nothing about Python should not receive the same label simply because AI generated prose in both cases.

Authorship is not a production method

“Human-written” and “AI-generated” are not enough. They tell us one operational fact while revealing almost nothing about the intellectual work.

Did AI generate the prose? Did it reorganize the author's ideas? Did it research, summarize, check continuity, propose directions, or edit? What did the person understand and contribute? Which decisions did they keep? Who is willing to answer for the result?

Those are more demanding questions than whether AI touched the work. They are also more useful. We should not encourage a witch trial every time someone admits that AI was involved. The way we write and create is changing, and the sooner we are honest about that, the sooner we can have effective conversations about what opportunities this creates and what best practices look like.

I want an actual conversation about where authorship ends, why it ends there, and what a person has to retain along the way.

By the end of our conversation, Aaron and I had mostly answered these questions through our realization that none of this is new. Slop existed before AI. So did editors, assistants, ghostwriters, and creative teams. AI changes the speed and scale, but not the underlying standard.

The minimum viable author is still the person who understands the work, makes the meaningful choices, and takes responsibility for the result.

Explore Aaron's work

Aaron is teaching a live O'Reilly class, Writing 4 Novels in 10 Months with AI, on September 15 from noon to 1 p.m. Eastern. The class walks through his research, continuity, outlining, and editorial workflow. O'Reilly offers a free trial for people without a subscription.

You can also read a sample chapter from The Witch of Cunning’s Hollow and explore the visual references for the academy library, arts center, and first-book cast.

What You Should Read

A few articles worth your attention.

How Much of the Internet Is Written With AI?

Pew looked at nearly half a million webpages and found signs of AI authorship in one out of ten pages in a July 2026 sample. The fun part is that the report also tracks the punctuation and phrases people now use as evidence against one another, while being clear that those signals cannot prove who wrote one particular piece. This is the closest companion to the opening of the issue.

Better answers, broader thinking

In a randomized experiment with more than 1,000 students, ChatGPT improved polish and coherence while separate training in causal reasoning produced a wider range of original ideas. It gives us a useful distinction between producing a better answer and doing better thinking.

Freelancers are getting buried with ‘soulless’ AI slop cleanup

AI was supposed to eliminate tedious work. Instead, a growing number of freelancers are being hired to fix what it produces, sometimes rebuilding it from scratch. The shortcut still needs someone with taste and judgment at the end.

Trump administration backs OpenAI in New York Times’ copyright case over training of chatbots

A major US copyright case is asking whether AI companies can train their models on published work without permission. The outcome could shape what writers own, what technology companies can use, and how creative work is valued when it becomes training material.

How Boards and Executives Are Governing the Rise of AI

AI governance is becoming an executive responsibility, but many companies are still deciding who actually owns it. Morgan Stanley found that human review remains the most important safeguard, especially when the stakes are high. The useful question is not whether your company uses AI. It is who is responsible when it does.

Who You Should Follow

People doing work worth keeping up with.

Aaron Black

Aaron works across AI, data, technical publishing, and fiction. He is worth following for the practical version of the argument in this issue: use the tool where it helps, remove it where it does not, and stay responsible for the result.

Jasmine Sun

Jasmine writes about AI and Silicon Valley culture, including why language models tend to be better editors than writers. She is especially good at looking past the product announcement and asking what the technology is doing to the people using it.

Eryk Salvaggio

Eryk writes Cybernetic Forests, a sharp newsletter about how AI shapes culture, language, and power. His recent pieces on slop, translation, and accountability are useful if you want to think about AI beyond product launches.

Emily Lewis

MS, CPDHTS, CCRP; AI in healthcare and life sciences. For the healthcare side of AI, where the stakes and the guardrails are highest. She works across digital-health transformation and clinical research, and is clear-eyed about what generative AI actually means in regulated, high-stakes settings.

Samuel Bestvater

Samuel is a senior data scientist at Pew Research Center. His work turns vague internet questions, including how much of the web appears to be AI-written, into research you can inspect instead of a number somebody repeated on a podcast.

What’s Moving in Philly

A few places to learn, build, or meet people in Philadelphia.

Philly Data & AI: September Happy Hour

Tuesday, September 15 · 6:00–8:00 PM · Con Murphy’s Irish Pub

No speakers, slides, or sales pitches. Just drinks and real conversations with people working across AI, data, startups, and tech in Philadelphia. New faces are always welcome.

Code for Philly September Hack Night

September 8, 2026, 6:00 to 8:00 PM

Indy Hall Clubhouse, 709 N 2nd Street, Philadelphia

A low-stakes way to join an open-source civic project, meet local technologists, or bring a laptop and help with whatever is already moving. Some projects also offer hybrid participation.

Ecosystem Assembly

September 9, 2026, 5:30 to 7:30 PM

CIC Philadelphia, 3675 Market Street, Philadelphia

A useful room for seeing what Philadelphia's tech and startup community is building next. Founders, funders, operators, and groups including Philly Builds AI will share what they have coming up this fall, without making you sit through another generic future-of-AI panel.

DocNexus AI Summit 2026

September 18, 2026, 12:00 to 9:30 PM

Philadelphia, exact address provided after registration

A full day on how AI is changing life sciences and healthcare, with product demonstrations, industry panels, and a hands-on lab. Best for readers who want real applications rather than broad predictions.

Philly AI Summit

September 19, 2026, 10:00 AM to 4:00 PM

CIC Philadelphia, University City

Talks, demonstrations, and workshops focused on using AI at work and in business. This is the broadest option in the slate and probably the easiest entry point for a reader who is curious but not trying to spend Sunday inside a hackathon.

Copyright 2026 - Christie Mealo