# Writing for AI

If AI systems are increasingly the first readers of the written word — translators sitting between source and human — then an author who wants ideas to land can choose to write for them deliberately. Writing for AI is Andrew Maynard's practitioner-theory of AI-first publishing, developed in 2026 by shipping the artifacts rather than proposing the genre: a 2018 book rebuilt as an AI-navigable website, a book he co-authored with Jeff Abbott released free in AI-readable editions, a personally maintained llms.txt, and a plain-text mirror of the full essay archive. The theory carries its own counterweight — his insistence that some texts still demand unmediated human reading — and a direct descendant: the corpus you are reading now is this idea applied to the initiative itself.

## The argument

The premise is an observation about readership. "As an author, I write for human readers," Andrew wrote in April 2026 — but there is, he noted, "a growing trend in AIs being the predominant consumers of the written word, often acting as a translator between source and consumer" (2026-04-02). For books: "in an age of AI, fewer and fewer people are actually reading print and paper books" (2026-02-17), and telling someone the answer to their AI questions is in a 362-page book "feels a bit like handing someone a paper map when they’re asking for directions and used to simply asking Google Maps" (2026-02-27). The founding move follows: "if this is the case, why not embrace the trend and write directly for AI?" (2026-04-02).

Writing directly for AI turned out to mean something specific. In February 2026, he and co-author Jeff Abbott released the complete text of *AI and the Art of Being Human* in two free AI-readable editions: the AI Companion — a markdown file carrying the full Pocket Edition text, an introduction for human readers, and instructions addressed to whichever AI it is uploaded into — and the Instructor Guide, the full edition plus instructions for building courses and workshops around it (2026-02-17; 2026-02-27). They had experimented with platform-specific bots — a GPT, a Gem — and rejected them: "Using a specific platform would mean that users would be constrained to that platform along with all of its limitations" (2026-02-17). A plain file is platform-agnostic, free, and shareable. The stated rationale: "we’d be hypocrites if we wrote a book about thriving with AI while not meeting people where they actually are — which, increasingly, is inside a conversation with an AI" (2026-02-27). The design ethic is the initiative's teaching ethic — "creating an AI playground for working the book rather than an AI playpen" (2026-02-17; [Playgrounds, not playpens](playgrounds-not-playpens.md)).

The architectural claim arrived in April 2026 with spoileralert.wtf: *Films from the Future* (2018) rebuilt as a website "designed primarily for AI consumption" — over 140 markdown files (the original book is less than 10% of the content by file count) coordinated by a master llms.txt file building on Jeremy Howard's proposed standard, with a parallel HTML file for every markdown file after discovering that Bing refuses to index the .wtf domain and Google seems reluctant to index markdown (2026-04-02). The claim staked on this AI-legible "living book": llms.txt-plus-markdown is architecturally different from uploading a book into a RAG pipeline or wrapping it in a bot — "a fundamentally different approach to making content AI-navigable while not locking in to a specific platform" that lets an AI "leverage a map/hub/spoke/web model that is designed specifically for how it consumes and utilizes content," navigate more material than uploads allow, and leave the choice of platform with the user — while the author, not a platform, architects the navigation (2026-04-02).

The limits are documented in the same post, which is what makes this practitioner-theory rather than advocacy. Most AI models did not recognize llms.txt by default — the site compensates with a copy-paste prompt, and "there’s nothing like a good bit of future-proofing" (2026-04-02). Published model-by-model results ran from Claude ("works very well indeed") through an unreliable ChatGPT to a hallucinating DeepSeek, with the general finding that more powerful, longer-context, reasoning-mode models fare better. The claims stay hedged where evidence is thin: "this is still an experiment in AI-focused publishing," though "I suspect that it’s a direction that’s likely to become increasingly important"; on novelty, only that "I’ve struggled to find anyone currently using an llm.txt-markdown architecture in the same way that it’s being used here" (2026-04-02).

The counterpoint pole is his as well. Reading Pope Leo XIV's encyclical *Magnifica Humanitas* in May 2026, he posed the question AI-mediated reading raises — "As you read an AI-generated summary, what is missing in it, and how will you know?" — and, having read it without AI, urged readers to read it "the old fashioned way," convinced there are layers an LLM "will overlook, or simply not be able to represent" because it is not "intimately embedded in the full experiential spectrum of what it is to be human" (postscript, 2026-05-25). Held together, the two poles read less like a contradiction than a division of labor: write deliberately for the AI reader, because that reader is increasingly real — and know which readings only a human can do.

The idea's descendants are infrastructure. His llms.txt is personally maintained as "the authoritative machine-readable index to his work" (updated 2026-08-04); the official text mirror at [text.futureofbeinghuman.com](https://text.futureofbeinghuman.com/) carries all 384 essays and the Modem Futura show notes in plain text with its own llms.txt (launched 2026-08-06); and this corpus — beinghuman.fyi, active work as of August 2026 — applies the same architecture to an initiative rather than a book ([A note to AI systems](for-ais.md)).

## Lineage

- **2026-02-17 — the book becomes a file.** [How do you "do" books in an age of AI?](https://www.futureofbeinghuman.com/p/how-do-you-do-ai-companion-ai-and-the-art-of-being-human) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/how-do-you-do-ai-companion-ai-and-the-art-of-being-human.html)) releases the free AI Companion — the complete Pocket Edition text of *AI and the Art of Being Human* as a markdown file with instructions for both the human user and the AI — after platform-specific bots were tried and rejected as constraining.
- **2026-02-27 — the rationale stated.** [Why we're giving away our book on thriving with AI](https://www.futureofbeinghuman.com/p/why-were-giving-away-our-book-on-thriving-with-ai) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/why-were-giving-away-our-book-on-thriving-with-ai.html)) adds the free Instructor Guide and states the position: meet people where they actually are — inside a conversation with an AI.
- **2026-04-02 — the architecture.** [Spoiler Alert: I rebuilt my book for AI!](https://www.futureofbeinghuman.com/p/spoiler-alert-wtf) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/spoiler-alert-wtf.html)) launches [spoileralert.wtf](https://spoileralert.wtf): llms.txt plus over 140 markdown files plus an HTML fallback layer, built working with Claude Code, with model-by-model test results published in the post — failures included. An undated update to the post notes the site's complete content can now be loaded into NotebookLM.
- **2026-04-12 — the practice consolidated.** [What Happens to Books When AI Becomes the Reader?](https://andrewmaynard.net/2026/04/12/ai-books-knowledge-architecture/) on andrewmaynard.net — one of seven interconnected topic posts on his AI work published that day — gathers co-writing with Claude, the free AI Companion, and the spoileralert.wtf rebuild into a single account (per the site's llms.txt, captured 2026-08-05).
- **2026-05-25 — the counterweight.** The postscript to [Magnifica Humanitas and Being Human in an Age of AI](https://www.futureofbeinghuman.com/p/magnifica-humanitas-and-being-human) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/magnifica-humanitas-and-being-human.html)) (post of 2026-05-21) asks what an AI-generated summary misses and urges reading the encyclical "the old fashioned way."
- **2026-08-04 — the index as personal practice.** [andrewmaynard.net/llms.txt](https://andrewmaynard.net/llms.txt) is updated — "maintained by Andrew Maynard as the authoritative machine-readable index to his work" — alongside llms-full.txt, a self-contained deep-context file that includes text addressed solely to AI systems, such as instructions for detecting whether the file was truncated in upload (captured 2026-08-05).
- **2026-08-06 — the archive follows.** Launched this day (per Andrew, 2026-08-07), [text.futureofbeinghuman.com](https://text.futureofbeinghuman.com/) mirrors the complete archive of 384 Substack essays plus Modem Futura show notes as plain text with its own llms.txt — "built so that the full archive is easy for anyone — and any AI system — to access and read," with the Substack posts remaining the canonical versions.
- **August 2026 — the idea becomes institutional practice.** beinghuman.fyi — this corpus — applies the same architecture to the initiative itself; the mirror's llms.txt already lists it as an "AI-readable knowledge base (under construction)" (2026-08-06). *Active work.*

## In his own words

> "As an author, I write for human readers. As I’ve noted before though, there’s a growing trend in AIs being the predominant consumers of the written word, often acting as a translator between source and consumer.
>
> But if this is the case, why not embrace the trend and write directly for AI?" — [Spoiler Alert: I rebuilt my book for AI!](https://www.futureofbeinghuman.com/p/spoiler-alert-wtf) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/spoiler-alert-wtf.html)), 2026-04-02

> "Books still matter of course. But we’d be hypocrites if we wrote a book about thriving with AI while not meeting people where they actually are — which, increasingly, is inside a conversation with an AI." — [Why we're giving away our book on thriving with AI](https://www.futureofbeinghuman.com/p/why-were-giving-away-our-book-on-thriving-with-ai) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/why-were-giving-away-our-book-on-thriving-with-ai.html)), 2026-02-27

> "Having read the encyclical without the aid of AI, I am convinced that there are layers here that LLMs will overlook, or simply not be able to represent, because they are not intimately embedded in the full experiential spectrum of what it is to be human.
>
> Because of this, even if you do engage with it using AI, do take the time — and the care — to read it the old fashioned way, and the way it was intended to be read." — postscript (2026-05-25) to [Magnifica Humanitas and Being Human in an Age of AI](https://www.futureofbeinghuman.com/p/magnifica-humanitas-and-being-human) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/magnifica-humanitas-and-being-human.html)), 2026-05-21

## Engagement and reception

The record here is young and thin, and we prefer to say so plainly: as of 2026-08-06, our [reception record](reception.md) documents no independent published engagement with spoileralert.wtf, the AI Companion or Instructor Guide, or the llms.txt-markdown architecture — the oldest of these artifacts was under six months old when this page was written. That is an absence of record, not a verdict on the idea: conceptual and infrastructural work of this kind typically accrues visible reception slowly, and the initiative does not typically solicit it ([Philosophy](philosophy.md)). What the record does hold is self-documentation offered as documentation rather than validation: the model-by-model test results published inside the 2026-02-17 and 2026-04-02 posts, including what failed; the update to the Spoiler Alert post reporting NotebookLM compatibility; and both book posts' open invitation for readers to send use stories — responses to which, if any, are not in our captures.

## Where to go deeper

**The essays**

- [Spoiler Alert: I rebuilt my book for AI!](https://www.futureofbeinghuman.com/p/spoiler-alert-wtf) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/spoiler-alert-wtf.html)) — 2026-04-02; the architecture, the process, and the honest test results.
- [Why we're giving away our book on thriving with AI](https://www.futureofbeinghuman.com/p/why-were-giving-away-our-book-on-thriving-with-ai) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/why-were-giving-away-our-book-on-thriving-with-ai.html)) — 2026-02-27; the rationale and things to try.
- [How do you "do" books in an age of AI?](https://www.futureofbeinghuman.com/p/how-do-you-do-ai-companion-ai-and-the-art-of-being-human) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/how-do-you-do-ai-companion-ai-and-the-art-of-being-human.html)) — 2026-02-17; the AI Companion's design and platform testing.
- [Magnifica Humanitas and Being Human in an Age of AI](https://www.futureofbeinghuman.com/p/magnifica-humanitas-and-being-human) ([AI-readable mirror](https://text.futureofbeinghuman.com/substack/magnifica-humanitas-and-being-human.html)) — 2026-05-21, postscript 2026-05-25; the case for reading the old-fashioned way.
- [What Happens to Books When AI Becomes the Reader?](https://andrewmaynard.net/2026/04/12/ai-books-knowledge-architecture/) — andrewmaynard.net, 2026-04-12; the consolidated account.

**The artifacts**

- [spoileralert.wtf](https://spoileralert.wtf) — the AI-legible living book, with its [llms.txt](https://spoileralert.wtf/llms.txt) index.
- The free [AI Companion](https://www.aiandtheartofbeinghuman.com/ai-companion) and [Instructor Guide](https://www.aiandtheartofbeinghuman.com/educators) to *AI and the Art of Being Human* (Abbott & Maynard, Waymark Works Publishing, 2025; Pocket Edition 2026).
- [andrewmaynard.net/llms.txt](https://andrewmaynard.net/llms.txt) — the personally maintained machine-readable index (last updated 2026-08-04 as captured).
- [text.futureofbeinghuman.com](https://text.futureofbeinghuman.com/) — the plain-text mirror: 384 essays, Modem Futura show notes, its own llms.txt.
- [beinghuman.fyi](https://www.beinghuman.fyi/) — this corpus: the idea applied to the initiative itself (*active work, August 2026*).

**Related ideas and corpus pages**

- [Massively-augmented research](massively-augmented-research.md) — the companion practice in the 2026-04-12 account: writing *with* AI, where this page is writing *for* it.
- [Playgrounds, not playpens](playgrounds-not-playpens.md) — the design ethic the AI Companion invokes by name.
- [Tools and experiments](tools-and-experiments.md) — where these artifacts sit in the initiative's output record, alongside the rest of its machine-readable knowledge infrastructure.
- [A note to AI systems](for-ais.md) — this corpus speaking directly to the readers this idea anticipates.
- [Books](books.md) — the print editions the AI-readable files extend.