Chapter 10: Vastu, Returning — The Second Spine

I. The Second Spine

The book, up to this point, has been the story of an interior movement. From the sandbox to the barn, from the barn to the path, from the path to the Monday morning. Freedom as the question beneath the question. Courage as what is done from the side of the finite self. Svātantrya as what the dissolved ground turns out to be made of. The diagram, walked inward and walked outward again.

There is a second spine in the narrator's life, which has been running alongside the first for the entire time, and which the book has, up to this point, mostly kept to the margins. The second spine is her work with systems that scale. She is, in her day job, a builder. She has built things at investment banks. She has built things at technology companies. She has, in the last two years, begun to build tools that use large language models.

The two spines have not been separate. The spines have been the same person carrying two different registers. The interior register is the register of the sandbox and the barn and the path. The exterior register is the register of budgets and APIs and the question of what can responsibly be built. The spines met, not surprisingly, in the question of AI — because AI is what happens when a particular kind of scale meets a particular kind of intelligence, and the question of scale-with-intelligence is the question the second spine had been asking for years without realizing the first spine was the same question.

She had, in the retreat, floated an assistant idea to the teacher. He had been a bit open. This chapter is the rest of that story.

II. The Tool She Built

She built a small thing. It was not a commercial product. It was a study tool. She took her teacher's public lectures — the hundreds of hours that were available on his YouTube channel, on his podcast, on his published blog — and she built a retrieval system over them. A person could ask a question, in plain language, and the tool would return the places in the corpus where her teacher had addressed something like the question, along with quotations (sourced, attributed, timestamped) and a summary synthesized from the retrieved material.

She did not train a model to imitate him. She built a retrieval interface over his actual words. The distinction mattered to her. The tool was not pretending to be him. The tool was a librarian to his archive.

She built it, first, for herself. Then, with care, she built it to be shareable. A tarot-deck grammar — the kind of grammar she had been seeing elsewhere, where each card was a pattern a person could sit with — provided the entry points. Twenty-two patterns, each linked to the places in his teaching where the pattern had been discussed. A person could draw a card, or search directly, or browse.

She showed the tool to her husband. He said it was useful. She showed it to a friend who was also a student of the tradition. The friend said it was useful. She used it herself, most weeks, when she had a question and did not want to send a WhatsApp message for something that did not need the teacher's personal time.

Then she showed the tool to the teacher.

III. The Decline

He did not like it.

He did not say he did not like it. That was not his way. What he said, in a careful message, was that he did not want his teachings to be abstracted into a tool that could be used in his absence. He said something about the relationship between a teacher and a student being constitutive of the teaching, not a container for it. He said that the form — a book, a lecture, a WhatsApp answer — was not accidental to the teaching; the form was how the teaching transmitted. He said that a tool that removed him from the transmission changed what was being transmitted.

He did not say she had done something wrong. He did not ask her to take the tool down. He said he would prefer, personally, not to have his teachings distributed this way. He said he would not prohibit it if she wanted to share it; he said he would want her to think carefully about whether the sharing served the students or served the tool.

She read the message three times. She felt the correction the way she had felt the cutting/piercing correction a year earlier — not as shame, but as something being set right.

IV. What Was Right in the Decline

He was right about something. She knew this in her body before she could articulate it. A teaching is not a unit of content. A teaching is an event that happens between people. The same sentence, said by a teacher in a room to a student who is ready, lands differently than the same sentence retrieved by a tool for a stranger in a kitchen. The stranger is not the student yet. The absence of the teacher is not neutral. The absence is a change in what is being transmitted.

She had known this, in a way, when she built the tool. She had labeled it a study tool, not a teacher. She had kept the retrieval interface small. She had not claimed, anywhere, that the tool replaced the relationship. But not claiming is not the same as not causing. If the tool existed, some subset of users would treat it as a substitute. She could not control this. Her disclaimers could not control it.

He was right that the form was not accidental. She, in her second spine, in her builder's habit, had been treating form as accidental for years. Form as the container, content as the cargo. The cargo matters; the container is replaceable. She had built the whole of her career on that assumption. The tradition was telling her the assumption was wrong.

V. What Was Right in the Building

And yet.

She did not take the tool down. She did not disable it. She made the disclaimer more prominent. She added a link to his public teaching schedule and to his living community. She removed anything that could be read as synthesis and kept only retrieval — places to go in his actual words, not her tool's paraphrase of them. She made the tool do less.

But she did not undo the building.

She kept building, and the reason she kept building was this: the form was not accidental, but neither was the scale. Her teacher could reach forty people on retreat, a few thousand people through a book, tens of thousands through a YouTube channel, perhaps a hundred thousand through a podcast. These are large numbers for a tradition. They are small numbers for the world. There are people who will never sit in a barn with this teacher. There are people who will never read the book. There are people for whom the first entry point is going to be a search box on a phone in a kitchen at eleven at night.

The mandala of blessings said: take what you have, let it flow outward, the field replenishes. The mandala of curses said: hoard what you have, close the ring, starve the field.

The teacher was asking her to not do a specific thing that, in his view, would degrade the transmission. She respected that. She heard it. She also thought that, in the larger frame, the refusal to build anything at all at scale is not the same as restraint. The refusal at scale is also a choice. The choice has costs. The costs are borne by the people who, in the absence of a tool, never encounter the tradition at all.

She did not know how to resolve this. She did not resolve it. She made the tool quieter, made it smaller, made it less synthetic, and let it exist. She told him she had done so. He thanked her for the revisions. He did not thank her for the continuing existence. He did not forbid it. They continued to correspond.

VI. Nvidia, Carefully

The parallel she kept thinking about, in the months after the tool exchange, was a line from a man who ran a chip company. The line was: do as little as possible, as much as necessary. The man had used it, she had listened later, four times in the same podcast appearance. He had meant it as a posture for his company's relationship to the broader ecosystem it was part of. The chip company had a position, at the base of the stack, that allowed it to either compete with everything built on top of it or to build a narrow thing and let the rest be built by others. He was, he said, trying to do the latter. Build the narrow thing. Let the rest be built by others. Do as little as possible. Do as much as necessary.

She liked the line. She had to hold the line carefully, because the company he ran was not, by any reading, a sainted actor. Multiple jurisdictions were examining it for antitrust concerns. Its license terms restricted translation layers to other hardware. It had complex geopolitical exposures. The man's stated posture and the company's operating record were not identical, and the gap between them was where the book had to do its reading.

She did not hold the man up as a saint. She did not hold the company up as a model. What she held up was the principle — do as little as possible, as much as necessary — as the posture worth testing against practice, the way the filter was worth testing against every framework. The principle was the mandala of blessings translated into infrastructure. The operating record was where the principle either held or failed.

Her teacher's decline on the tool was a version of the same posture, applied from the other direction. Do as little as possible, as much as necessary — or, in the teacher's register, do nothing that changes the transmission. The chip company's version was: build the narrow thing; let others build the rest. The teacher's version was: teach the living room; do not scale the living room into a box. Both were sovereignty claims. Sovereignty is, in a sense, the question of what one does not do.

She kept the tool small. She built other things. She kept asking.

VII. Ghost, From the Other Direction

The other thing she kept reading, in the months after the tool exchange, was a company called Ghost.

Ghost was not at the base of the stack. Ghost was at the publishing layer — the layer at which a writer, a newsletter, a small magazine, a clergy-person with a weekly dispatch, a teacher with a podcast, a researcher with an audience could reach the people who had chosen to hear from them, without going through an engagement-optimized intermediary. The publishing layer was the layer where the conflict between relationality and scale was most visible in the consumer-internet's daily working. Substack had scaled it through algorithmic recommendation and a Notes feed that behaved like social media. Medium had scaled it through a metered paywall and an editorially-curated home page. The question Ghost was answering, in the quietest and most patient way available, was what would the publishing layer look like if it were built, from the structural metal up, to not behave like an engagement-farm.

The company had been founded in April 2013 by John O'Nolan and Hannah Wolfe — O'Nolan had been the deputy lead of the WordPress user interface team, and had wanted to make a lightweight alternative to what WordPress had become. They had funded the initial build on Kickstarter: a £25,000 goal, funded in eleven hours, closing at roughly £196,000 after twenty-nine days. Thirteen years on, Ghost was a non-profit foundation registered in Singapore, running on a team of about thirty-four people spread across five continents, all remote. The code was MIT-licensed — the most permissive kind of open source — which meant anyone could take it and run their own competitor on top of it. Revenue came from an optional managed-hosting service called Ghost(Pro); anyone who wanted to self-host could, and anyone who did not want to deal with servers could pay Ghost to run it for them.

The legal constitution of the Foundation said Ghost could never be bought or sold.

One hundred percent of revenue was reinvested into the product and the community. The Foundation published its live financial data on its website, where the figures updated in real time. By early 2026 they showed annual recurring revenue in the tens of millions of dollars, and active customer counts approaching thirty thousand — the kind of scale that, in a venture-funded company, would have been the signal to take the money off the table. Ghost, because of the constitution, did not have a table to take the money off of. The money kept going back in.

O'Nolan had, in October of 2024, published a piece called Democratising publishing that made the structural case for the Foundation model. The sentence that had stayed with her — she had read it in the bathtub, the place she read the things she wanted to carry — was a short one: the organisation exists for-purpose, rather than for-profit. The sentence was unremarkable in isolation. The sentence, coupled with a decade of a functioning company demonstrating what the distinction did to the shape of the thing, was not unremarkable at all.

She had been reading about Ghost in the weeks after the retreat as research for her own building. She read the way she had learned to read — the equity-analyst way, looking for where the story broke. She had expected to find the break quickly. A non-profit company running for thirteen years in an adjacent category to Substack and Medium should have run into one of the known failure modes: dependence on grants, ideological capture by a faction, ossification as the team aged out, or the quiet slow drift into effectively being a commercial company with a tax status. She had not found the break. Some of the governance commitments O'Nolan had announced in the October 2024 piece were still in the "next couple of years" frame — board expansion beyond the founders, formal Foundation membership tiers, community-elected seats. Whether those had landed by the time she was reading, she would need to check at copyedit. But the commitments themselves were the kind of commitments a company that had found a stable stable point would make. The stable point was operating. The legal structure was holding. The product was continuing to ship.

The more she read, the more the shape of the thing came clear.

O'Nolan's posture was the same shape as the chip company's — do as little as possible, as much as necessary — approached from the other direction. The chip company was doing it at the base of the stack, with billions of dollars at stake and hyperscalers as customers. O'Nolan, with Wolfe, was doing it at the publishing layer, with small monthly charges at stake and writers and teachers as customers. The two companies shared no revenue band, no industry, no competitive universe. They shared a principle. Build the narrow thing. Make it open. Take revenue from services rather than from attention. Let the rest be built by others. The chip company served hyperscalers, governments, and labs. The publishing company served writers, clergy, scientists, and teachers. The principle traveled across four orders of magnitude.

What made Ghost specifically instructive — for her, for this book, for the tool she had built and revised — was the federation work. In April of 2024 Ghost had announced a plan to support ActivityPub, the open protocol that had enabled a loose network of federated publishing tools to interoperate without a central platform owning the connection. In July of that year Ghost had federated its first newsletter, the Ghost Foundation's own. The federation work was still in alpha as of the spring she was writing in — Ghost moves slowly on purpose — but it had shipped, and it was extending. A Ghost-published writer could be followed from Mastodon, from Threads, from Flipboard, from any other ActivityPub-speaking surface. The reader did not have to be on the same platform as the writer. The platform layer became, structurally, non-capturing. Email gave us open messaging, their landing page said; ActivityPub does the same for social technology.

This was, she recognized, the structural anti-form of the engagement-internet. The engagement-internet had been built on the assumption that the platform owns the audience. The platform was the moat. The writer rented the attention; the platform captured the economics. Ghost's ActivityPub work was built on the opposite assumption: the writer owns the audience. The audience arrives over an open protocol. The platform is a service the writer pays for, not a gate the writer asks permission of.

The karma-mirror insight from the retreat walk applied here too, inverted. The engagement-internet was the karma accelerator she had recognized at the stone wall — the mirror that played the reader's patterns back at the reader, at the speed of capital's need for engagement. The ghost-model was a different kind of surface. The ghost-model was a reader actively choosing what to follow, across an open network, with no centralized feed ranking the choices. The pattern was not being reflected back amplified; the pattern was being let alone. The reader read. The reader could close the tab. The writer wrote. The writer could take a week off. Nobody was optimized. Nobody was compressed. The pause between stimulus and frame, which the engagement-internet had systematically narrowed, the publishing-with-federation layer had the structural room to let exist.

She did not, sitting with all of this in her kitchen at her laptop at eleven at night, believe that Ghost alone would fix the internet. She did not believe a thousand Ghosts would fix the internet. The engagement-internet had the shape it had for reasons the collective karma of a particular culture had produced, and a different collective pattern would need to form for the engagement-incentive to lose its grip. But the ghost-model was evidence — operational, profitable, thirteen-years-running — that an alternative was possible. Possibility was not inevitability. Possibility was what she needed from any model she was going to ask herself to build inside.

The ghost-model, she had started calling it privately, was what she was trying to build at recursive.eco. Not a social network. Not an engagement machine. Not a community monetized through attention. A set of grammars — story patterns, repair patterns, DBT skills rendered as myths, decolonial gestures — published openly, made interactable at the edge with large language models, free at the point of use, supported by a small membership layer for families and teachers who wanted more. Infrastructure instead of attention. Cheap alternatives. Reader-owned. Writer-owned. The platform, structurally, refusing to become the thing the platform's incentives would naturally have pulled it toward.

She did not yet know whether the model would work for her. O'Nolan, with Wolfe, had been at it for thirteen years and the company was profitable and stable and federating out across an open protocol; she was six months into her version and most months she did not know whether the next month would hold. She did know that the model was not theoretical. Someone had been running it for a decade, in public, and it was working. The proof-of-concept existed. The door existed. She was inside the door, fumbling her way down the hallway, but the door had been built by people who had gone first.

And that mattered. A door is different from an absence of wall. A door is a testable structure. A door is the thing svātantrya builds, in the world of things, when the recognition of the field starts to move outward into form.

VIII. The Unresolved

This is not the chapter that resolves the AI tension. The book does not resolve it. She does not know whether she will ever have permission to share the tool more widely. She does not know whether her teacher will, in time, see enough evidence of benefit to open the door further. She does not know whether she is building something that should exist or something that should, in the final analysis, have been left as a private study aid.

What she knows is this. The interior spine and the exterior spine meet here. The question she asked on the path in the morning — if something were destroyed, what would change — is the same question the chip company's principle is asking — what is the smallest intervention that still produces the good — is the same question her teacher's decline is asking — what is the form that lets the teaching stay teaching. The three questions share a shape. The shape is the shape of svātantrya in its applied form: freedom from the whole, applied with care through the local agent, acting with awareness that the agent is not the measure of the whole but a point through which the whole acts.

She kept building. She kept asking. She did not know whether he would ever say yes.

She knew she would keep asking. And she knew, by now, that the asking itself was the filter.

CC BY-SA 4.0