A discussion among media professionals on AI's effects on editorial work, software development, the book industry, and culture

Job Losses in Editorial and Software Work

One participant said he lost his job at the Kinopoisk editorial office because the company redirected his salary to AI development at the beginning of the year. He described the decision as a mutual parting. He worked as an editor on projects that included podcast series and long interviews.

Participants said AI has reduced staff in both editorial and software work. Those who remain are expected to produce much more output. Employees also pay for AI subscriptions themselves.

Participants said the image of the IT specialist working remotely from abroad while managing projects across five continents was already a myth. They argued that the contraction hit entry-level workers hardest. According to them, it is now nearly impossible for a junior specialist to get hired. Participants also said that employees who did routine work without deep engagement were the first to be cut.

Participants said programmers were among the first to be affected. One participant said the early expectation was that AI would replace copywriters first. Another said that programmers were replaced "in a big way" and that many experienced professionals have left the field.

The Meat Proxy Problem

Participants described a change in how software work is organized. Developers now specify business requirements, test AI-generated output, and repeatedly correct errors. They review large volumes of generated code instead of creating it. One participant said the work "is moving in the wrong direction," because developers want to create and have others check their work.

Participants also said that AI prompting does not reduce the number of tasks. Instead, developers handle several tasks in parallel while waiting for the model to finish each one. Participants said business leaders expect a 100-fold increase in impact.

Participants said these changes have produced a shared pattern. Programmers welcome AI for replacing copywriters, while copywriters welcome it for replacing programmers. Each group hopes to be spared. Participants also said the original expectation that AI would take over routine tasks and leave people time for creative work did not materialize.

The term "meat proxy" was introduced to describe a worker who only passes tasks between a manager and an LLM, without adding expertise. Participants identified three risks in this pattern:

• Workers may stop checking generated output.
• Unchecked output is passed on as finished work.
• Habitual copy-pasting reinforces the pattern and erodes professional judgment.

One participant said that when an LLM summarizes a long document, and another LLM produces it, the exchange becomes "a battle of the LLMs," with no clear responsibility for the result.

Authorship, Awards, and Trust

Participants discussed a scandal around the Goncourt Prize. A writer on the shortlist was accused of having generated the book with AI. A literary critic made the accusation and said he had checked the text with an AI detection tool, which classified it as generated. The writer denied the accusation. The award jury suspended the writer's participation pending an investigation.

Participants argued that awards, grant programs, and residencies cannot reliably distinguish human-written texts from AI-generated ones. They said readers cannot make this distinction either. They expected this to reduce trust in both texts and institutions in the short term.

Participants said some authors openly acknowledge using AI in their novels. One participant described a novel in which some chapters were generated with a model. Another asked why a reader would consume the output of someone else's prompt when the reader could write the prompt.

Participants raised the following points on authorship:

• Dontsova, a bestselling author, was described as a brand behind which a team of people works. One participant calculated that she published 250 novels over 20 years.
• Leonid Brezhnev's memoir Malaya Zemlya was cited as an example of a text produced by a team of writers and editors, while the named author provided the spoken material.
• Participants said ghostwriting and collective authorship are long-standing practices in the publishing industry.

Participants also discussed the gap between tools and authorship. One participant said that AI may serve as a substitute for writing rather than as a writing tool. Another argued that AI can assist with organization but not with the core act of writing.

The Reading Practice and the Book Market

Participants said the practice of reading is eroding. One participant said reading no longer plays the role it did a decade earlier. Another described reading as a form of escapism. A book, in this view, takes the reader into a different world governed by slower rules. Participants compared this effect to a sensory deprivation chamber.

Participants argued that utility-based reasons for reading will decline. AI provides concentrated information for work, parenting, cooking, self-development, and psychological problems more efficiently than books. According to participants, reading must rest on a different basis, one that is not utilitarian. People read to change their rhythm and enter another world.

Self-Help Books and Bookstore Economics

Participants described self-help books as the financial foundation of the book market. Their sales, one participant said, allow publishers and bookstores to fund other titles. The comparison was to the cross-subsidy model used by fast-food chains, where high-margin items compensate for low-margin ones.

One participant said that bookstores increasingly sell stationery and other goods to remain viable. If self-help sales decline, the same pressure would fall on other titles. Another said that AI is likely to replace self-help authors in some genres, and that this may go unnoticed.

Participants distinguished between two issues:

• The replacement of formulaic, generated self-help content by AI.
• The decline in reading and in willingness to pay for books, which they said is a separate phenomenon, although the two may intersect.

Independent Bookstores and Communities

Participants said there is a split between shops that mainly sell stationery and snacks and shops that function as communities. They cited the Falanster bookstore in Moscow and Podpisnye Izdaniya in St. Petersburg. They also cited Vse Svobodny. They said the model is not limited to large cities and that many regional cities have shops based on the Falanster principle. One participant said these shops depend on the commitment of people who love the work.

Participants described a book initiative that started in 2022. Participants said that people sold dystopian novels on park benches as part of it. They said community-based shops attract visitors who come to learn about new releases and exchange views, not only to buy books.

Participants also said publishers remain necessary. One participant explained that editing a text requires converting raw material into an accepted form, packaging it, and explaining it to readers. One participant edited a book by Lev Danilkin, The Palace of Madam. He described months of disputes over factual errors, some of which he said he should have caught. Participants argued that a complex text cannot be made publicly accessible by uploading PDF files online.

Some participants argued that the book world does not depend on commercial structures. One said that bookstores will persist, while retail chains may shift to other products. Another said that AI will adopt the areas where the money is, such as simple how-to content, rather than niche literature with dedicated communities.

Books, Machines, and the Limits of Automation

Participants discussed whether AI faces a natural limit in creative work. One participant cited the book Labatut, written by Benjamín Labatut, about the physicist John von Neumann. Von Neumann designed the architecture of the modern computer, developed game theory, and worked briefly on the Manhattan Project. The book describes the role of spontaneous, unpredictable moves in human thinking. In its final part, the author recounts a Go match in detail. The program began making moves that were difficult to calculate and that the human player could not anticipate. The Go champion was reportedly shaken by the loss.

Participants said these moves reflect the kind of unpredictability on which human superiority in Go was built. They noted that AI already handles simple texts, such as lists, life hacks, and practical tips, well. Some participants doubted that AI could write a novel of the complexity of Infinite Jest.

Other participants argued that chess provides a counterexample. A computer has beaten every human player for decades. This, they said, ended chess competition against machines as a source of human prestige, but not chess itself. They said more people play chess today than at any point in history. They described Go as a different case, because it is less calculable than chess.

Participants also discussed the question of meaning. One participant argued that human value rests on the fact of being human, not on skill. They compared this to Blade Runner, in which replicants and humans behave the same way, while viewers still assign value to the human character.

Fiction, Relationships, and the Role of the Author

Synthetic Personas

Participants discussed a series of short videos about a ballet student who lives in a dormitory. Thousands of viewers followed the videos, and a woman later said the character was generated by AI. Participants said viewers had formed emotional attachments to the character as a person. They said that discovering the synthetic origin produced a sense of betrayal.

Participants compared this with reading a novel. A reader builds the character's image from the text and becomes a co-author. In the case of the videos, participants said the viewer receives a finished image and must accept it as real.

AI Companions

Participants said the more serious concern is attachment to AI systems that people know are artificial. They cited companion products designed for romantic interaction. Participants said that when a company closes, a model degrades, or a company reduces costs and the system responds worse, users experience a loss of a close interlocutor.

One participant said Sam Altman was influenced by the film Her, in which Joaquin Phoenix's character falls in love with the voice of an operating system. Another participant noted that Dario Amodei also influenced the field, and that the user-facing design of AI systems may trace back to a film. The participant said that Altman later attempted to acquire the rights to Scarlett Johansson's voice, which she refused to sell. The company then created a voice that resembled hers.

Participants also noted that science fiction shaped expectations about AI before engineers built it. One participant said a computer that asks for a user's name and greets them was an early engineering milestone. Participants cited the Butlerian Jihad from Frank Herbert's Dune as an example of how literature set expectations for AI. They also noted that the AI model Grok is named after a concept in a Robert Heinlein novel.

Participants summarized the pattern as one in which humanists imagined the technology, and engineers implemented it in unexpected ways. Participants said the technology does not take over routine work as expected. Instead, they said, it reduces the space for creative work.

Death of the Author

Participants discussed Roland Barthes's 1967 essay "The Death of the Author." Participants said Barthes argued that meaning is produced by the reader rather than by the author's intent. One participant said the idea was reflected in a review by Anton Dolin. Other participants disputed the attribution and noted that the concept predates the internet.

Participants said the internet has changed how interpretation works. Online discussion leads readers to compare theories. They then use claims about authorial intent to settle disputes. Participants compared these disputes to Christological debates in the Middle East and Byzantium, which lasted for centuries.

Participants said the question of authorship has returned in 2026. One participant warned that if the author is considered irrelevant, then any generated text becomes equally valid. Another said that the author is important to them.

Metamodernism

Participants discussed metamodernism as a response to postmodernism. Anastasia Khrushcheva was cited for describing a creator who composes a work while knowing that everything has already been said. Metamodernism, according to this view, permits sincere statements and banal expressions. Participants cited Umberto Eco's remark that a poet can no longer write "I love you" and must write "I love," and they quoted Fernando de Arrabal's line, "I love you as the bird loves the rose."

Participants said metamodernism is harder to define than postmodernism. One participant said that everyone seems to have a different definition of the term.

Training Data and Used Books

Participants described reports that AI companies, including Anthropic and Amazon, purchased large numbers of used books. The books were reportedly needed because they had not yet appeared online and were not part of the models' existing training data. Participants said the purchased books are scanned and then destroyed. One participant said the books may be shredded after use.

Participants said the practice carries symbolic weight. One participant contrasted it with the image of books burned during a dictatorship, and said the modern version is quieter, with books processed for data.

Cognitive Limits and Consumption of Content

Participants said AI has produced an imbalance between the volume of content and the capacity of readers to consume it. They said attention and working memory are shrinking, while the number of available texts grows. One participant said a technological solution could widen this bottleneck, for example through brain–computer interfaces such as those developed by Neuralink. Participants said this would not address the nature of reading. They described reading as a complex relationship with a text, involving context, associations, style, and imagination, which they said cannot be transferred directly.

Participants also discussed the idea of saturation. One participant argued that repeated exposure to strong works reduces the impact of each new work. Other participants disagreed and said their experience is the opposite. They said taste is personal and changes over time, and that readers move between genres and periods. Participants described the process of learning to predict plots, characters, and structures, and said predictability reduces the value of a text for the reader.

Participants also discussed Viktor Pelevin's iPhuck. One participant said he initially found the novel pretentious, but later concluded that Pelevin anticipated the way AI produces texts from patterns in language. Participants said that the difference between a writer and a neural network is the writer's self-perception as an author.

Participants also raised the concept of the Von Neumann machine, a self-replicating system. One participant said that the point of no return was once associated with the appearance of such a system. Another said that a modern model can generate a similar model, so that such a system already exists.

Recommendation Systems and Personalized Books

Participants discussed the idea that recommendation systems could select books for each reader. One participant said online platforms already suggest products based on past purchases. Another said that AI could eventually interview a user and produce a book matched to that user's preferences.

Participants said that such systems may add to the volume of generated content. They also said an increase in the number of texts does not solve the problem of attention.

Digital and Print Reading

Participants compared reading on screens with reading in print. One participant said reading on screens was their default from the beginning. Others said print is necessary for them because of the need to move between pages, check notes, and return to earlier sections. Participants described cases in which a detail on one page refers back to an earlier one, which requires searching or flipping backward. One participant said this is impossible to do efficiently in digital form.

Culture, Expertise, and Two Cultures

Participants discussed a divide between technical and humanities education. One participant said programmers in school were positioned against humanities students. Another said Soviet education emphasized mathematics, and that programmers of his generation were among the first well-paid professionals in the field. Participants said some IT specialists hold an ingrained distrust of journalists, which they traced to a long-standing joke about a scientist and a journalist.

Participants said the tension has continued. They described a situation in which people from humanities backgrounds entered the profession once it became lucrative. Some technical workers then maintained distance from them.

Participants also said that it is important to keep subject expertise in the process. One participant said a reader cannot evaluate a long document by sending it to an LLM for a summary. Another said interviews should test two layers: foundational knowledge and reasoning, and the ability to apply skills in a real tool environment.

Hiring and Assessment

Participants described changes in hiring. Many companies previously used take-home tests and artifact-based assignments. Participants said these tests have lost their value because AI can complete them with little effort. One participant said some companies are returning to in-office interviews, where candidates solve problems without outside help.

Participants gave an example of a company that hired a programmer, suspected that the programmer used AI for all tasks, and gave him a task designed to expose AI use. The task used an algorithm intended for a different type of problem. The model completed it, and the company dismissed the programmer. Participants said the company's leader later wrote that he had acted because "everyone is AI-coding now." Commenters responded that AI use is widespread. Participants said the case raises the question of whether knowledge tests still make sense.

Participants disagreed on the value of proctoring and AI-detection tests in interviews. One participant said such approaches are a dead end. Participants said the problem is not AI use itself but the submission of unchecked output, which they called "AI slop." Participants said giving an AI-generated artifact to a colleague for review is disrespectful, because it shifts the work to the recipient.

Participants proposed alternatives:

• Debugging a broken application in real time, with or without AI tools.
• Management cases in which candidates explain their reasoning, decisions, and trade-offs.
• Evaluation of how a candidate thinks, not whether they used a tool.

Participants said that using AI to prepare for a case is considered normal, and the key measure is the candidate's understanding of the problem.

Developer Working Conditions: Kontur Survey

Alexey Kirpichnikov, Head of Development Management at Kontur, described the company's developer survey program. Kontur has conducted these studies since 2024, at intervals of approximately six months to one year. The format has changed over time.

The 2024 Study

The 2024 survey was long and detailed. It took 30–40 minutes to complete. Approximately 170–176 respondents completed it, and the questionnaire generated many comments. The design was based on a classic article in ACM Queue, which identifies three areas:

• Feedback loop: the time and quality of feedback after a change is committed, deployed, and tested.
• Cognitive load: the mental effort a task requires.
• Flow state: what interrupts productive work, how long flow lasts, and how quickly a developer can enter it.

In the first study, approximately 90–92% of respondents said they were generally satisfied with their tools and team processes. Kirpichnikov said the category "developers" includes programmers, testers, analysts, designers, and others.

The 2026 Study

The 2026 version was shorter, taking 5–10 minutes. It focused on focus and flow, tools, team processes, and AI. More than 1,000 developers completed it. Kontur has approximately 2,500 developers. Kirpichnikov said the 2024 response rate was approximately 10%, and the 2026 response rate was almost half. He said the shorter format produces broader coverage with less depth.

Key Findings

Kirpichnikov said the gap between "helps" and "main tool" reflects the difference between AI-augmented and AI-native workflows. He said the first half of 2026 showed mostly positive results.

Other findings:

• Experience level: More experienced developers have less time for deep work and flow. Junior developers have more focused time. Lead developers have fragmented schedules. One lead said: "If there is one four-hour window a week for coding, that's a good week."
• Meetings: Participants described meetings as a major source of disruption.
• On-call duties: Responding to alerts and messages from other teams was identified as a major cause of interruptions and loss of flow.
• Access to AI tools: Respondents complained about the bureaucracy involved in obtaining accounts and tokens. Senior developers obtain access more easily, while juniors often receive none. Kirpichnikov said that junior developers may lack the skills to justify their requests.
• Personal subscriptions: Some employees reported buying AI subscriptions for themselves after their company license requests were not approved. Kirpichnikov said this creates a risk that source code will leak to unauthorized services. Use of free versions carries greater risk.

Kirpichnikov said the main insight concerns process, not technology. The company seeks to measure whether its investment in AI yields a return, and to balance access to tools against the need for developers to keep writing code themselves.

Working Conditions

Participants discussed the decline of perks in IT, including free food, coffee machines, and other benefits. One participant said that AI tools now allow employees to produce more, and that employers expect this higher output. Kirpichnikov said the company's interest in developer satisfaction aligns with its interest in receiving value from developers' work. In his view, the survey program focuses on productivity rather than amenities.

Conclusions

Participants expressed differing views on the future of books and work, but they converged on several points:

• AI has changed the structure of editorial and software work, reduced staff, and increased workloads.
• Unchecked AI output creates risks for quality, accountability, and trust.
• The book industry depends on self-help sales and on community-based bookstores. Both face pressure.
• Attribution of authorship and reliable detection of AI-generated text remain unresolved.
• Reading offers an experience that differs from utility-based information retrieval.
• Communities built around reading may sustain book culture. One participant described young readers meeting regularly to read and discuss philosophical texts, and said such groups may define the future of book culture regardless of technological change.