My July 2025 study establishing that literary agents almost never read submissions for quality drew some notoriety and even animosity, but a surprising result was that a few people who worked in publishing reached out to thank me.
I wrote literary fiction with dark humour and I didn’t even bother querying, going straight to self publish. It’ll reach whoever it needs to and if that’s no one well, my friends read it and that’s cool.
I will need to reread this a few times for absorption. Do we know what we don’t know yet? That’s the start of addressing how to move forward. One major problem we have in indie/self publishing is the disorganized space of millions of authors. Pitchforks and torches against the curtain wall of trad and Amazon enshittification platforms. We can lament this in thousands of words but at some point will we start to undermine the battlements with a unified strategy? Are we there yet? Great article, BTW.
This piece describes ian industry that has been making literary fiction financially viable through institutional subsidy — prestige, awards, critical infrastructure, marketing budgets — and is now concluding the subsidy isn't worth the return. The question of whether literary fiction can survive without institutions is actually the more interesting one. It survived for centuries before institutions claimed it.
There's so much realism in Michael Church's "2028 is When Traditional Publishing Plans to..." that I will only say to other writers, "Ignore this Substack post at your own peril." My wife (I married extremely well for a Minnesota farmboy) and I published our first books a little more than fifty years ago. We were lucky to find our first agent when he came to us. Those first works led to enough success we could leave the workaday world behind and make a good (increasingly good, after her first appearance on the NYT list) living for five decades.
The life of working writers could be tough, even with success behind you. In the ffty years we labored in the vineyards of the Lord, we have seen so many changes that it is difficult to count them. When hardback imprints were absorbed into the borg of the Big Five, back in the 90s, we survived. When paperback distribution was consolidated into the hands of a few jobbers, we still made a living. When the publishing world went woke, as Church describes, we got out.
So his characterization of today's commercial publishing is spot on, so far as we are concerned. Agents are a vanishing breed of gatekeepers. Commercial publishing has to work too hard at starting new trends to make the Big Five a realistic bet. If a writer is not financially independent and able to bankroll a book marketing plan, the odds of making a hit out of even a solid manuscript are, as Church so charmingly puts it, "Clap Ass Zero."
Going independent is a viable option but the writer should be a very shrewd marketer, as well as a great stylist and a deep-dive researcher. AI can help. And yes, I'm saying that AI has a place in writing, just as Church describes the ways publishers are using AI while publicly decrying the dehumanization of the new robotic world.
Church's 2025 stem-winder is even more pertinent today. Dig it out and use it to push yourself as a writer into the real world. Somehow, the scribbling trade seems too complicated today. But don't give up. If you have the overwhelming urge to scribble, you have no choice, even when success seems distant. I think it always was a long shot, but it's the only thing I know how to do. So excuse me, I have to go finish the rewrite of abook that started growing in my mind in 1972. I don't know whether it will ever be a success but I know I can't wait to see how the story turns out.
And Thanks to another scribbler, James Ross of the North, for pointing me toward the Church teachings.
Thank you... 30 years after studying literature... 20 years as a librarian (public library) and writer (3 novels out)... the main conclusion for me is... the literary market does not give a shit about literature... and a culmination in 2028 sounds rather reasonable to me
But what to do about the AI slop flooding the market? I think there should be a label “This work is AI generated”. It’s now so easy to ‘create’ a book, people jump in and if they can even sell a small number of the hundreds they release, they deem it worth it. I’ll stick with small presses and hope they survive.
Small presses are good, but I'm not sure you need an agent for that. People who sell nothing would be delighted with even a handful of paid readers. There should be a way or business model (and there probably is) that is more than happy to seek out small subsets of readers for small-time writers. Publishing is dead; just look at the poor quality of books today versus what they were thirty years ago. Remember the old Penguin Classics paperbacks? Those were a delight to read. Now they're little more than cheap newsprint with a cheap cover. Almost every book I pick up at a bookstore is just garbage quality-wise. That's not the sign of a thriving business. It's the sign of a dying business.
In practice, I think you need an agent for small presses. I have a friend who submitted direct and had to wait 500+ days. That's not getting read.
Unfortunately, publishing runs on relationships. You really need someone to ensure your work is fairly and completely read, to get a timely decision, and to figure out for you, if the decision was a rejection, why it was what it was, and what to do next.
A good agent keeps editors honest. The problem is that the agents who can do that for you are few, and inaccessible. You don't query them. They take two or three clients per year, mostly on referrals. Hence, the need for an agent-for-agents, which is ideally your family publicist—if you haven't got one, trade probably isn't for you.
It's unscalable and destined to produce mediocrity, but it's controllable and therefore provides reliable income for the people it benefits. The fear a lot of people have in the writing world is that, when the institutions finally collapse, we'll have just chaos.
It may not be good for people's livelihoods. On the other hand, it will at least be interesting.
It can produce better income if agents and presses acted like sales reps instead of literary critics. We don’t need literary critics, we need people who can sell. Everything’s mediocre. I have no idea why we’ve come to hate that about ourselves. Let posterity decide who among us is great; I’d like to get supplemental income.
Not only are publishers scamming, but the entire university system that has turned to MFA programs and writing/producing/illustration courses, and the residencies markets to generate income streams the way sports commercialization has. Making bets on successful writing? Sure, why not. All life is a gamble. The first hurdle is to get read ... whether or not you get a review. The universities are still the gatekeepers, and there's no denying that DEI and other forms of Marxist conformity are required for one to have success as a writing academic for these programs, courses, conferences. Schmoozing is mandatory --- the very thing that writers abhor because they're WRITING! For example, I'm not a 23-year-old, social media addicted, gorgeous BIPOC, divergent, lesbian, recovering from fill-in-the-blank, but I tell some great stories. Naturally, this makes me marked for automatic failure. BUT .... I know history, and I know how wrong, wrong people can be about the public. Trump wasn't supposed to be elected, remember? And Britain was going to kick the ass of the Nazis, but needed American help to do so. Go figure.
Because writing is different. It’s publishing that’s no different from any other industry. The shock probably comes from the slap of that reality after having been fed the dream of the artist whose writing is discovered by an editor who finds the work so valuable and relevant or simply beautiful that it must be distributed to the masses (ala Maxwell Perkins and his authors).
Well Kristen this is what they want us to believe. "The dream" keeps the wheels greased, from submission fees to retreats to writing classes to paid editors to coaches and on and on. Just like NASA tells kids we can all be astronauts. No, we can't.
BTW nothing's changed. All Perkins's writers were insiders and ferocious self-promoters
So what’s going to happen to all the money-spinning MFA programs that only exist because they’re seen as the ticket out of the slush pile for literary fiction (and as employment for people with MFAs)?
This all makes me wonder how realistic is the idea of "an online bookstore whose recommendation system, unlike Amazon’s, uses no social-proof signals because full-text reading gives it all the information it needs". You previously noted (in a comment to the prior post I'm quoting) that this is an open research question, given the need to both check for literary quality and at the same protect against people trying to game the system. But even if the research question were solved it seems that the practicality of such a service would depend on the economics of doing LLM-assisted evaluation.
For example, as I understand it, today it would cost on the order of $1 to check a 100,000-word novel for AI writing using a commercial service like Pangram. Any online bookstore like this would presumably attract lots of (no doubt AI-generated) submissions that would need to be evaluated and rejected, so it seems the cost of evaluation would need to be to be much lower in order to make a service like this profitable, unless you charge would-be authors a per-submission fee to cover the cost. Otherwise if you have to evaluate 1,000 or 10,000 works to find one good enough to publish, and the typical published work sells only 1,000 copies on average, then you can't spend more than a few cents on evaluating each submission.
You raise some good questions. AI detection is a messy problem, and I don't think anyone has cracked it. False positives happen all the time, and we don't have a good model of when to call something AI—if it was copyedited by AI, is it AI? We'll punt on that by saying it's not that important; if we can read for quality, then people whose quality bars exceed what AI can achieve will not be recommended AI-generated content.
That assumes, of course, that we can get good at reading for quality. Perfect? No. Better than literary agents? Feasible. We're replacing a flawed signal with a less flawed, but still flawed, signal.
I wouldn't use a frontier model. I'd start with a smaller language model that either has hierarchical sparse attention or can be put in a framework to perform it, so we can fit a whole book in the (extended) context window. We're talking about 10B parameters—smaller than GPT-3.5, and far smaller than today's frontier models. Since it's small, we can post-train it to specialize on literary recognition. We don't need the ~10T parameters of frontier models because we don't need the generality they possess; we only need the model to do one thing.
We personalize. There's no such thing as an objective quality score for a book. What we get instead would be a list ("vector") of scores that correspond, loosely, to different types of readers (or genres) and their impressions. (This could be manually forced, or it could emerge in training—probably better to let training find them, especially if there are going to be hundreds or thousands.) Maybe there are 400 of them. A score (or dimension) that corresponds to mystery readers who want lyrical prose, one that corresponds to pulp readers who like profanity, one that corresponds to Christian readers who don't, etc. We publish the model. We don't publish personalization; everyone's weight vector is private.
Thus, let's say—for simplicity's sake—there are three dimensions (x, y, z) that correspond, I don't know, to: writing quality, sexual content, and mystery. Anyone can feed their novel into a local copy of the neural network and find out, say, that their novel scores at (5, 15, 6). Let's say I like spicy but easy books, so my preference vector is (1, 5, -3). You like lyrical cozy mysteries so your preference-weight vector is (4, -5, 4). The algorithm scores that novel at 1x + 5y - 3z = 57 for me and 4x - 5y + 4z = -31 for you. It's far more likely to serve it to me than to you.
By publishing the model, you achieve transparency. There's nothing hidden or fishy in the algorithm itself. By keeping personalization private, you prevent people from running gradient ascent (adversarial attack) to produce superslop that beats the algorithm but is actually junk—the same method as is used to create pictures of cats that neural networks recognize as dogs.
This design isn't perfect, but it's cheap. Using a smaller language model allows you to keep costs down. It'd still take millions of dollars to do this right, but the costs are in data acquisition—you need data to train the model on—not compute itself, and so you can keep inference costs very low: probably a few cents per book at most.
Michael, thank you for a wonderfully informative piece. Like many others, I can't stop thinking about it, either. Even if I didn't have a debut literary novel that I'm preparing to query, it would still feel like a gut-punch.
Thanks to Frank for his comment about an online bookstore. Michael, it seems you've been giving this a great deal of thought. So I'll put this question to you, Frank, and everyone else here who has commented: what does an alternative to Amazon looks, feels, and functions like?
(I love your X,Y,Z coordinates concept. Love it!) A search on Claude for self-publishing platforms besides Amazon brings up a fairly short list with Apple, Google Play, Kobo, and retailers like B&N. When I asked Claude a clarifying question about retailers like B&N accepting self-published titles directly from authors, this was part of Claude's response:
"The storefronts generally don't label a book as "self-published" — to a browsing reader, it just looks like any other book. The practical difference is in discoverability: traditionally published books tend to get better algorithmic placement, editorial curation spots ("Best of the Month" picks, etc.), and marketing support. Self-published titles have to earn visibility through sales rank, reviews, and sometimes paid promotion."
This makes me love the transparent X,Y,Z concept even more. How much money do you think would be necessary for a start-up? Seriously. How much?
The bad news: Startup costs are probably in the millions—$2-4 million, probably. This is a pittance by tech-startup standards, but it's prohibitively expensive for a new entrant to publishing that publishers would rather not exist (since it demystifies their supposedly impeccable selection process) and about which Amazon will be ambivalent.
The technology isn't the expensive part. A solid grad student in computer science with machine learning experience can build everything you need, at least on the backend. Getting data is the hard part. You need a few thousand book-length manuscripts to train on. To convince people that you can sift through slush piles to find publishable manuscripts, you need to be able to show that traditionally published books (labeled examples) were spotted in slush piles. You need empirical validations and quantitative results, not, "This picks better books than trade, because I looked."
Gaining trust and reach is an even harder problem. You need buyers to know you exist. You need writers to know that you're self-hosting your LLMs, because they justifiably don't want their manuscripts to end up on Sam Altman's hard drive, where their work might appear in future training sets. I'd also avoid AWS out of vigilance. The probability that that Amazon would screw over a small AWS customer to protect its book business is very low—it cares far more about web services than books these days—but you gain credibility if you can say you use your own hardware. This reduces your long-term costs, but it increases your upfront personnel costs, because there are thousands of rent-a-devs who know how to build on top of AWS, but very few people who know how to set up and scale hardware.
The software can be built in a couple weeks. You can do the job with commodity hardware (GPUs) if you can find someone who'll set everything up. Gaining users and trust while building the platform is going to cost about a million per year for at least two years. Then you'll have a proof of concept.
Here's a long-term problem. If you succeed, you'll almost certainly get an offer from Amazon to buy you for a few hundred million. Your Silicon Valley investors will expect you to take it. If you don't, they will probably fire you and replace you with someone who will. After that, it's out of your control how this is used. Amazon might take a hands-off approach toward the quirky product that picks books than existing algorithms and gatekeepers. It might not; it could destroy the product at publishers' behest.
It can be done. There are other ways. But it is hard to fund this as a conventional business when it is so reliant on others' trust, in addition to labeled data. You need a six-figure budget just to get the world to know your product exists, let alone trust it.
I don't value their signal, nor would I respect anyone who does, but the fact is that you need one to get anywhere in traditional publishing.
Editors can't read everything, so they pushed the job downward unto a new class of intermediaries. They mostly don't read either, but publishing can deflect blame. Can't get a literary agent to read you? You must be a terrible writer, then. No, we're not going to check and see if the system's wrong. Go away and figure it out.
FUK. I knew it. Literary fiction with minor weirdness is my favorite thing to write, but I was suspecting that Lit fiction is being assassinated. I will sprout wings and fly to the moon before one living being accepts free ARC books for literary fiction.
I cannot stop thinking about this. I know I am revealing bias and predilection. But your essay only seems to confirms long held suspicions. Your apparent insider knowledge doesn’t make me joyful for the confirmation…just saddened. This is censorship through power. Hopefully your writing exposes the nakedness of this particular emperor at scale.
I wrote literary fiction with dark humour and I didn’t even bother querying, going straight to self publish. It’ll reach whoever it needs to and if that’s no one well, my friends read it and that’s cool.
I will need to reread this a few times for absorption. Do we know what we don’t know yet? That’s the start of addressing how to move forward. One major problem we have in indie/self publishing is the disorganized space of millions of authors. Pitchforks and torches against the curtain wall of trad and Amazon enshittification platforms. We can lament this in thousands of words but at some point will we start to undermine the battlements with a unified strategy? Are we there yet? Great article, BTW.
This piece describes ian industry that has been making literary fiction financially viable through institutional subsidy — prestige, awards, critical infrastructure, marketing budgets — and is now concluding the subsidy isn't worth the return. The question of whether literary fiction can survive without institutions is actually the more interesting one. It survived for centuries before institutions claimed it.
Hear, hear.
There's so much realism in Michael Church's "2028 is When Traditional Publishing Plans to..." that I will only say to other writers, "Ignore this Substack post at your own peril." My wife (I married extremely well for a Minnesota farmboy) and I published our first books a little more than fifty years ago. We were lucky to find our first agent when he came to us. Those first works led to enough success we could leave the workaday world behind and make a good (increasingly good, after her first appearance on the NYT list) living for five decades.
The life of working writers could be tough, even with success behind you. In the ffty years we labored in the vineyards of the Lord, we have seen so many changes that it is difficult to count them. When hardback imprints were absorbed into the borg of the Big Five, back in the 90s, we survived. When paperback distribution was consolidated into the hands of a few jobbers, we still made a living. When the publishing world went woke, as Church describes, we got out.
So his characterization of today's commercial publishing is spot on, so far as we are concerned. Agents are a vanishing breed of gatekeepers. Commercial publishing has to work too hard at starting new trends to make the Big Five a realistic bet. If a writer is not financially independent and able to bankroll a book marketing plan, the odds of making a hit out of even a solid manuscript are, as Church so charmingly puts it, "Clap Ass Zero."
Going independent is a viable option but the writer should be a very shrewd marketer, as well as a great stylist and a deep-dive researcher. AI can help. And yes, I'm saying that AI has a place in writing, just as Church describes the ways publishers are using AI while publicly decrying the dehumanization of the new robotic world.
Church's 2025 stem-winder is even more pertinent today. Dig it out and use it to push yourself as a writer into the real world. Somehow, the scribbling trade seems too complicated today. But don't give up. If you have the overwhelming urge to scribble, you have no choice, even when success seems distant. I think it always was a long shot, but it's the only thing I know how to do. So excuse me, I have to go finish the rewrite of abook that started growing in my mind in 1972. I don't know whether it will ever be a success but I know I can't wait to see how the story turns out.
And Thanks to another scribbler, James Ross of the North, for pointing me toward the Church teachings.
Thank you... 30 years after studying literature... 20 years as a librarian (public library) and writer (3 novels out)... the main conclusion for me is... the literary market does not give a shit about literature... and a culmination in 2028 sounds rather reasonable to me
Self-publishing is the future.
But what to do about the AI slop flooding the market? I think there should be a label “This work is AI generated”. It’s now so easy to ‘create’ a book, people jump in and if they can even sell a small number of the hundreds they release, they deem it worth it. I’ll stick with small presses and hope they survive.
Small presses are good, but I'm not sure you need an agent for that. People who sell nothing would be delighted with even a handful of paid readers. There should be a way or business model (and there probably is) that is more than happy to seek out small subsets of readers for small-time writers. Publishing is dead; just look at the poor quality of books today versus what they were thirty years ago. Remember the old Penguin Classics paperbacks? Those were a delight to read. Now they're little more than cheap newsprint with a cheap cover. Almost every book I pick up at a bookstore is just garbage quality-wise. That's not the sign of a thriving business. It's the sign of a dying business.
In practice, I think you need an agent for small presses. I have a friend who submitted direct and had to wait 500+ days. That's not getting read.
Unfortunately, publishing runs on relationships. You really need someone to ensure your work is fairly and completely read, to get a timely decision, and to figure out for you, if the decision was a rejection, why it was what it was, and what to do next.
A good agent keeps editors honest. The problem is that the agents who can do that for you are few, and inaccessible. You don't query them. They take two or three clients per year, mostly on referrals. Hence, the need for an agent-for-agents, which is ideally your family publicist—if you haven't got one, trade probably isn't for you.
You describe a terrible business model. Fortunately, that business model is dying.
It's unscalable and destined to produce mediocrity, but it's controllable and therefore provides reliable income for the people it benefits. The fear a lot of people have in the writing world is that, when the institutions finally collapse, we'll have just chaos.
It may not be good for people's livelihoods. On the other hand, it will at least be interesting.
It can produce better income if agents and presses acted like sales reps instead of literary critics. We don’t need literary critics, we need people who can sell. Everything’s mediocre. I have no idea why we’ve come to hate that about ourselves. Let posterity decide who among us is great; I’d like to get supplemental income.
Not only are publishers scamming, but the entire university system that has turned to MFA programs and writing/producing/illustration courses, and the residencies markets to generate income streams the way sports commercialization has. Making bets on successful writing? Sure, why not. All life is a gamble. The first hurdle is to get read ... whether or not you get a review. The universities are still the gatekeepers, and there's no denying that DEI and other forms of Marxist conformity are required for one to have success as a writing academic for these programs, courses, conferences. Schmoozing is mandatory --- the very thing that writers abhor because they're WRITING! For example, I'm not a 23-year-old, social media addicted, gorgeous BIPOC, divergent, lesbian, recovering from fill-in-the-blank, but I tell some great stories. Naturally, this makes me marked for automatic failure. BUT .... I know history, and I know how wrong, wrong people can be about the public. Trump wasn't supposed to be elected, remember? And Britain was going to kick the ass of the Nazis, but needed American help to do so. Go figure.
Writing's no different than any other industry. Why is this so shocking, or disappointing, to so many?
Because it’s a creative endeavor, so my wishful thinking is that it’s immune.
Because writing is different. It’s publishing that’s no different from any other industry. The shock probably comes from the slap of that reality after having been fed the dream of the artist whose writing is discovered by an editor who finds the work so valuable and relevant or simply beautiful that it must be distributed to the masses (ala Maxwell Perkins and his authors).
Well Kristen this is what they want us to believe. "The dream" keeps the wheels greased, from submission fees to retreats to writing classes to paid editors to coaches and on and on. Just like NASA tells kids we can all be astronauts. No, we can't.
BTW nothing's changed. All Perkins's writers were insiders and ferocious self-promoters
So what’s going to happen to all the money-spinning MFA programs that only exist because they’re seen as the ticket out of the slush pile for literary fiction (and as employment for people with MFAs)?
This all makes me wonder how realistic is the idea of "an online bookstore whose recommendation system, unlike Amazon’s, uses no social-proof signals because full-text reading gives it all the information it needs". You previously noted (in a comment to the prior post I'm quoting) that this is an open research question, given the need to both check for literary quality and at the same protect against people trying to game the system. But even if the research question were solved it seems that the practicality of such a service would depend on the economics of doing LLM-assisted evaluation.
For example, as I understand it, today it would cost on the order of $1 to check a 100,000-word novel for AI writing using a commercial service like Pangram. Any online bookstore like this would presumably attract lots of (no doubt AI-generated) submissions that would need to be evaluated and rejected, so it seems the cost of evaluation would need to be to be much lower in order to make a service like this profitable, unless you charge would-be authors a per-submission fee to cover the cost. Otherwise if you have to evaluate 1,000 or 10,000 works to find one good enough to publish, and the typical published work sells only 1,000 copies on average, then you can't spend more than a few cents on evaluating each submission.
You raise some good questions. AI detection is a messy problem, and I don't think anyone has cracked it. False positives happen all the time, and we don't have a good model of when to call something AI—if it was copyedited by AI, is it AI? We'll punt on that by saying it's not that important; if we can read for quality, then people whose quality bars exceed what AI can achieve will not be recommended AI-generated content.
That assumes, of course, that we can get good at reading for quality. Perfect? No. Better than literary agents? Feasible. We're replacing a flawed signal with a less flawed, but still flawed, signal.
I wouldn't use a frontier model. I'd start with a smaller language model that either has hierarchical sparse attention or can be put in a framework to perform it, so we can fit a whole book in the (extended) context window. We're talking about 10B parameters—smaller than GPT-3.5, and far smaller than today's frontier models. Since it's small, we can post-train it to specialize on literary recognition. We don't need the ~10T parameters of frontier models because we don't need the generality they possess; we only need the model to do one thing.
We personalize. There's no such thing as an objective quality score for a book. What we get instead would be a list ("vector") of scores that correspond, loosely, to different types of readers (or genres) and their impressions. (This could be manually forced, or it could emerge in training—probably better to let training find them, especially if there are going to be hundreds or thousands.) Maybe there are 400 of them. A score (or dimension) that corresponds to mystery readers who want lyrical prose, one that corresponds to pulp readers who like profanity, one that corresponds to Christian readers who don't, etc. We publish the model. We don't publish personalization; everyone's weight vector is private.
Thus, let's say—for simplicity's sake—there are three dimensions (x, y, z) that correspond, I don't know, to: writing quality, sexual content, and mystery. Anyone can feed their novel into a local copy of the neural network and find out, say, that their novel scores at (5, 15, 6). Let's say I like spicy but easy books, so my preference vector is (1, 5, -3). You like lyrical cozy mysteries so your preference-weight vector is (4, -5, 4). The algorithm scores that novel at 1x + 5y - 3z = 57 for me and 4x - 5y + 4z = -31 for you. It's far more likely to serve it to me than to you.
By publishing the model, you achieve transparency. There's nothing hidden or fishy in the algorithm itself. By keeping personalization private, you prevent people from running gradient ascent (adversarial attack) to produce superslop that beats the algorithm but is actually junk—the same method as is used to create pictures of cats that neural networks recognize as dogs.
This design isn't perfect, but it's cheap. Using a smaller language model allows you to keep costs down. It'd still take millions of dollars to do this right, but the costs are in data acquisition—you need data to train the model on—not compute itself, and so you can keep inference costs very low: probably a few cents per book at most.
Michael, thank you for a wonderfully informative piece. Like many others, I can't stop thinking about it, either. Even if I didn't have a debut literary novel that I'm preparing to query, it would still feel like a gut-punch.
Thanks to Frank for his comment about an online bookstore. Michael, it seems you've been giving this a great deal of thought. So I'll put this question to you, Frank, and everyone else here who has commented: what does an alternative to Amazon looks, feels, and functions like?
(I love your X,Y,Z coordinates concept. Love it!) A search on Claude for self-publishing platforms besides Amazon brings up a fairly short list with Apple, Google Play, Kobo, and retailers like B&N. When I asked Claude a clarifying question about retailers like B&N accepting self-published titles directly from authors, this was part of Claude's response:
"The storefronts generally don't label a book as "self-published" — to a browsing reader, it just looks like any other book. The practical difference is in discoverability: traditionally published books tend to get better algorithmic placement, editorial curation spots ("Best of the Month" picks, etc.), and marketing support. Self-published titles have to earn visibility through sales rank, reviews, and sometimes paid promotion."
This makes me love the transparent X,Y,Z concept even more. How much money do you think would be necessary for a start-up? Seriously. How much?
The bad news: Startup costs are probably in the millions—$2-4 million, probably. This is a pittance by tech-startup standards, but it's prohibitively expensive for a new entrant to publishing that publishers would rather not exist (since it demystifies their supposedly impeccable selection process) and about which Amazon will be ambivalent.
The technology isn't the expensive part. A solid grad student in computer science with machine learning experience can build everything you need, at least on the backend. Getting data is the hard part. You need a few thousand book-length manuscripts to train on. To convince people that you can sift through slush piles to find publishable manuscripts, you need to be able to show that traditionally published books (labeled examples) were spotted in slush piles. You need empirical validations and quantitative results, not, "This picks better books than trade, because I looked."
Gaining trust and reach is an even harder problem. You need buyers to know you exist. You need writers to know that you're self-hosting your LLMs, because they justifiably don't want their manuscripts to end up on Sam Altman's hard drive, where their work might appear in future training sets. I'd also avoid AWS out of vigilance. The probability that that Amazon would screw over a small AWS customer to protect its book business is very low—it cares far more about web services than books these days—but you gain credibility if you can say you use your own hardware. This reduces your long-term costs, but it increases your upfront personnel costs, because there are thousands of rent-a-devs who know how to build on top of AWS, but very few people who know how to set up and scale hardware.
The software can be built in a couple weeks. You can do the job with commodity hardware (GPUs) if you can find someone who'll set everything up. Gaining users and trust while building the platform is going to cost about a million per year for at least two years. Then you'll have a proof of concept.
Here's a long-term problem. If you succeed, you'll almost certainly get an offer from Amazon to buy you for a few hundred million. Your Silicon Valley investors will expect you to take it. If you don't, they will probably fire you and replace you with someone who will. After that, it's out of your control how this is used. Amazon might take a hands-off approach toward the quirky product that picks books than existing algorithms and gatekeepers. It might not; it could destroy the product at publishers' behest.
It can be done. There are other ways. But it is hard to fund this as a conventional business when it is so reliant on others' trust, in addition to labeled data. You need a six-figure budget just to get the world to know your product exists, let alone trust it.
Guess I'll write.
Ain't that the truth.
What about small independent presses?
Self publish. I read LOTS of epic fantasy but I buy it direct from writers. Not from publishers and not from retailers.
This is a great point. Readers need to support indie writers, and in many cases, they are.
As a lit fiction author querying my debut novel, this made me break out the wine.
"literary agents' - there's your first problem.
I don't value their signal, nor would I respect anyone who does, but the fact is that you need one to get anywhere in traditional publishing.
Editors can't read everything, so they pushed the job downward unto a new class of intermediaries. They mostly don't read either, but publishing can deflect blame. Can't get a literary agent to read you? You must be a terrible writer, then. No, we're not going to check and see if the system's wrong. Go away and figure it out.
I agree, you need agents to get into trad-pub. With that in mind, I can only advise, don’t get into trad-pub.
FUK. I knew it. Literary fiction with minor weirdness is my favorite thing to write, but I was suspecting that Lit fiction is being assassinated. I will sprout wings and fly to the moon before one living being accepts free ARC books for literary fiction.
Whoa...
I cannot stop thinking about this. I know I am revealing bias and predilection. But your essay only seems to confirms long held suspicions. Your apparent insider knowledge doesn’t make me joyful for the confirmation…just saddened. This is censorship through power. Hopefully your writing exposes the nakedness of this particular emperor at scale.
Let’s hitch a ride on the Death Star?