In the Beginning Was the Prompt

I have been reading scripture again. Not the comforting kind. I had settled in on a Saturday morning with the particular smugness of a woman who believes she has earned a quiet weekend, and within forty minutes I had read a transcript of an Atlantic podcast that has cost me my faith - or, more accurately, has revealed that I have been holding two incompatible faiths simultaneously for some months now, and one of them is the kind of heresy you cannot undo just by closing the tab.

A Crisis of Faith, Conducted Quietly, on a Saturday

The podcast is called Galaxy Brain. The episode is an interview with one Max Spero, founder of a company called Pangram Labs, which exists for the sole and stated purpose of detecting AI-generated writing on the open internet. Mr Spero, who refers to himself in the interview with the dry self-deprecation of someone in too deep to be funny about it, calls himself a “slop janitor.” This is the first sign that I am dealing with a prophet.

I read the transcript with rising anxiety. By the time I reached the section in which Pangram’s detection software was used to determine that the Pope’s own social media team appears to have used AI to draft tweets warning about the dangers of AI, I had stopped reading and started pacing. By the time I reached the discussion of how Pangram is being adopted by publishers, academic integrity panels, and platforms across the internet, I had begun mentally drafting my resignation letter from my own job.

Because here is the thing. I am, professionally, a woman who teaches public servants and university staff how to use Microsoft Copilot to summarise their emails. I do this for a living. I have a slide deck. I have a reputation. I have, on occasion, been described in writing as “the AI person.” And what I have apparently been doing - what I have been doing with both hands, twice a week, for most of a financial year - is leading earnest professionals out of the wilderness of their own laboured prose and directly into the welcoming arms of an enormous, expensively trained, increasingly accurate machine that exists to identify their writing as morally compromised and have it flagged at the document level.

I have, in short, been ordained in a religion whose central rite is now considered, by another religion of equal standing, an act of grave moral transgression. And both religions are real. And both have congregations. And the congregations overlap by approximately one hundred per cent.

On the Founding of the First Church (The Hedonist Communion)

The first church was founded, as most modern denominations are, by accident and at scale. It is the Church of the Hedonists, and its theology is admirably simple: the Machine writes well; thou shalt let it. Its scripture was drafted in late November 2022, distributed by Sam Altman, and authenticated by an entire generation of LinkedIn users who, having once written posts in their own clipped, slightly underconfident voices, suddenly began producing rolling cathedral sentences full of em dashes and “in today’s fast-paced world” preambles, as though the whole platform had been overtaken by the Holy Spirit during a productivity workshop.

The Hedonists believe - with a faith I find, in private moments, almost touching - that the use of the Machine is a kind of grace. There is no email too short to summon it, no birthday card too personal, no condolence note too solemn. The Machine has a vocabulary larger than yours; therefore the Machine should write. The Machine has read more business books than you ever will; therefore the Machine should compose your strategic plan. The Machine, like the God of certain Calvinist sects, is doing the work whether you are aware of it or not. You may as well stop fighting and let it.

The problem with this theology is not strictly that it produces what Spero, in his Atlantic interview, refers to with surgical precision as “slop.” The problem is that the slop is everywhere. According to a survey cited by Spero in that same interview, around thirty-five per cent of newly published websites in 2025 were AI-generated or AI-assisted. Other estimates discussed in the episode suggest that as much as half of all new articles on the open internet are now coming out of large language models. This is not a denomination. This is a global majority church. It has more attendees than Catholicism. It has fewer rituals, but the rituals are easier to perform on a phone.

I was, I am ashamed to admit, in attendance. I was not just in attendance; I was passing the offering plate. Every workshop I have run in 2025 has, at some point, included a slide titled Let Copilot help you draft a quick event description. I have stood in front of three hundred people across the Territory and personally led them in their first synthetic email. Every one of those emails is, this morning, somewhere on Earth, capable of being identified as AI-assisted at a false positive rate of one in ten thousand.

One in ten thousand. That number has not stopped echoing in my skull since I read it. It is, statistically, a more reliable mechanism than any I personally use to determine whether the milk has gone off.

On the Counter-Reformation (The Church of Luddite, and Its Sacred Detection Engine)

Where there is a church there must, eventually, be a counter-reformation, and the counter-reformation has, mercifully, arrived. Its first prophet is Spero. Its sacred technology is a machine learning classifier built by training on millions of documents, then asking the major large language models to produce “synthetic mirrors” of those documents at scale, then learning the difference between the two. Its central revelation is that large language models, when asked to write, make what Spero calls a kind of mode collapse: a narrowing of all possible sentences down to a much smaller decision tree than any human would naturally walk. The Machine, it turns out, has its preferred paths. It prefers them very strongly. The Detector knows them.

This is the second church. Its name, I have decided, is the Church of Luddite, though I want to be precise: I do not mean the original nineteenth-century Luddites, who had a coherent labour-economic critique and were broken-hand cottage weavers being immiserated by the power loom. I mean the modern devotional movement, whose adherents wear the name with a small ironic smile, whose true creed is not anti-technology but anti-AI-technology specifically, and whose first commandment - articulated in seven thousand LinkedIn posts and a great many op-eds in respected newspapers - is this:

This is sacred knowledge. It is, in the gnostic sense, the hidden wisdom that separates the elect from the damned. The Luddite knows that AI hallucinates. The Hedonist does not. The Luddite has read the Pangram website. The Hedonist has not. The Luddite carries this knowledge in a small leather-bound notebook of mental bookmarks that they will, given any pretext at all, deploy in meetings.

The Luddite is, on the substance, mostly correct. This is what makes them so unbearable. There is no faster way to be wrong about an unbearable person than to discover, on closer inspection, that they have been right the entire time. They have read the studies. They have noted the citations. They have a screenshot, and they will produce it. And while you are dealing with the screenshot, they will quietly install the Pangram browser extension, which will run silently in the background of their Substack feed, awarding each post a synthetic-content score, and which will, by the end of the calendar week, have radicalised them further.

I downloaded the extension on Saturday evening. I scored my own LinkedIn feed. I will not be sharing the result.

In Which Even the Vatican Is Sloppy

Let me linger on the Pope. Not theologically - I have neither the standing nor the wardrobe - but as a parable.

Here is what the Pontifex incident, as reported by Wired and confirmed using Pangram’s detection software, tells us. The Pope, the highest moral authority of one of the world’s largest religions, employs a social media team. The social media team, plausibly young, plausibly time-poor, plausibly under deadline pressure from a comms calendar that does not pause for prayer, has been using AI tools to draft tweets - including, in at least one documented case, a tweet warning about the dangers of AI. The tweets are then published from an account read by tens of millions of the faithful. Pangram identifies several of these tweets as having been written with significant AI assistance. The Vatican declines to comment, which I believe to be the correct theological response.

This is what I propose to call, going forward, the Pontifex Loop, and I believe it to be the central condition of professional life in 2026. We are all, at every level of every institution, simultaneously denouncing and using the thing. The Pope’s social media team is using ChatGPT to warn against ChatGPT. The university communications office is using Copilot to draft the email asking lecturers to be careful with Copilot. The Vice-Chancellor is using Gemini to review the AI policy that prohibits the use of Gemini in formal review. We do not have a coherent theology. We have a comms team and a deadline.

I do not raise this to mock. I raise it because it is the precondition of everything that follows. If even the Holy See cannot keep the slop out of its own scripture, what hope does an MVIT educator in the Territory have, exactly, of holding the line in a one-day workshop attended by thirty professionals whose CIO has signed an enterprise contract worth more than my annual salary?

A New Decalogue, Roughly Engraved

What is the AI trainer to do? Caught between the two churches, expected by the Hedonists to deliver workshops in which I cheerfully demonstrate how to get Copilot to write a meeting agenda, and expected by the Luddites to spend the second half of the same workshop showing the same room how to identify and shame people who do exactly that - I have decided, in the absence of any institutional guidance whatsoever, to write my own commandments.

I have not gone up a mountain. I have not had a mount, nor a cave, nor a desert wilderness, nor any of the geographic features traditionally associated with the activity. I have gone, instead, into the kitchen on a Sunday afternoon at thirty-two degrees with a flat white going cold beside me, which is, I would argue, the modern Australian equivalent. The tablets are not stone. They are a Word document in which the autosave has, twice, generated a Copilot suggestion I have had to reject manually, with the spiritual reluctance of someone being offered a cigarette while attempting to quit. There are ten of them. They are, like all decalogues, both inadequate and not negotiable.

I am aware these are insufficient. I am aware that within thirty seconds of publication, a Hedonist will write to inform me that I have missed the spirit of the technology, and a Luddite will write to inform me that I have not gone nearly far enough. I welcome both letters. I will respond to neither, on the grounds that any reply will inevitably be drafted with at least a little Machine assistance, and the resulting hypocrisy will damage my standing with both readers simultaneously.

The Trainer’s Predicament (or, I Am the Reluctant Bishop of an Apostate See)

Here is the actual question, the one that produced the Saturday morning crisis, the one I have not yet answered.

I run workshops. The workshops have a structure. The first half is called Foundations, in which we cover policy, privacy, data sovereignty, and the various ways in which the Machine is, on a quiet Tuesday afternoon, perfectly capable of leaking your client’s name into a model owned by a company headquartered in another hemisphere. The second half is called Practical Use, in which I demonstrate, with appropriate caveats, how to get Copilot to draft a meeting summary, an event description, a polite chase-up email to a colleague who has not responded to the previous chase-up email.

I now have to decide whether to add fifteen minutes to the second half. The fifteen minutes would consist of opening the Pangram browser extension on the screen behind me, running the meeting summary I have just helped them draft through it, and reporting the verdict back to the room.

The lesson, in its current draft, runs roughly as follows. Right, everyone, here is how to get Copilot to write your community event description. Excellent. Wonderful. You all did very well. And now — buckle up — here is the moral revelation. The text you have just produced, with my full encouragement, in an MVIT short course you paid for, will be flagged as ninety-three per cent likely AI by the detection software your professional association has just signed an institutional licence with. To cleanse yourself of the impure machine-generated artefact, here is a tool that will publicly shame you for not having written it yourself. Does anyone have any questions? No? Lovely. The catering is out the back.

I cannot, in conscience, deliver that workshop. I also cannot, in conscience, deliver the previous workshop, in which I send people back to their desks with a tool whose outputs are now functionally radioactive in any context where a Pangram subscription has been purchased. There is no third workshop. There is only the gap between the two, into which I have fallen, and where I have been writing, on and off, for several weekends now.

Where the Theology Briefly Cracks

I will drop the costume here. Just for a paragraph. I will put it back on.

I do not know what to teach. I genuinely do not. I can teach Copilot. I can teach Pangram. I cannot teach both in the same room without watching twenty-three professional adults experience the same micro-aneurysm in real time, as they grasp that the productivity tool their organisation just paid for has a corresponding shame industry that has also been paid for, by other organisations, and that the two are converging on their inbox at roughly the same speed. I am, in this sense, less an AI literacy trainer than an interfaith chaplain at a stalemate. I cannot reconcile the two churches because the two churches do not want to be reconciled. They are, as the user manuals of the AI age, a single contradiction with a UI. And it is my job to stand in front of it on a Tuesday and pretend, for ninety minutes, that the contradiction is a learning opportunity.

Concluding Heresies

So. I have, for my own peace, declared myself excommunicated from both denominations. I am not a Hedonist. I write my own emails, including the one I sent yesterday morning that contained a typo I am still annoyed about. I am not a Luddite either. I will not stand at the front of a room and refuse to demonstrate a tool that the people in the room have been told, by their Chief Information Officer, will be central to their working lives within eighteen months. I am, instead, what every priest of every minor heretical sect throughout history has been: a woman with too many books and not enough authority, building a private liturgy out of contradictions and hoping no one important attends the service.

The Laterite Highway, in the dry, is long enough that you can have an entire crisis of faith between two roadhouses and arrive at the next one with a different theology than you left Pandanus Reach with. This is roughly what has happened to me. I left as a Hedonist. I am arriving, slowly, as something else — call it a Reformed Pragmatist, call it a Confused Anglican, call it the Bishop of the Discrepancy File. The point is that no one is in charge, the rainbow ball is still in the rear-view mirror, and the slop continues to rise across the internet at a rate that will, by the time my next workshop runs, be measured in scripture.

I keep my own commandments. I am not certain they cover anything. But I have written them down, which is - historically - how most religions have started, and how a great many have failed to end.