How The Walker Briefing Is Made

The architecture of public-facing analytical writing has shifted, and the shift carries a question for any writer who works inside the new conditions, namely what such conditions ask of them. The question is not optional neither does it answer itself.
What I have to settle, in writing under those conditions, is what my own practice requires. The conditions are not external to the writing. They run through it and shape what the relationship between writer and reader can hold. That is what this article addresses.

There is a proximate occasion as well as the structural one already named. In Australia, the academic and editorial response infrastructure for AI use is being built faster than any consensus on what it is for. At Australian Catholic University, detection tools produced false-positive rates of two to five per cent. Against a student population measured in the tens of thousands, the consequences accumulate rapidly. That is the kind of structural condition the response infrastructure is now operating in: a response being assembled while the practice it answers continues to change underneath it. To write public-facing analytical work in those conditions and to leave the matter unsettled is no longer compatible with my own standards.

Walker Briefing is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

The answer this article will give is operational. What follows specifies, in the production of this publication, what AI does, what I do, and where the line is drawn between them. I publish on Substack under the name of the Walker Briefing. You may be curious to know how I create the articles. Other authors have their own style and I will not assume any one standard of writing. The article is about how I create the Briefing and not about any other writing.
There develops a relationship between writer and reader and that relationship rests on the material produced, something that remains a mystery to readers and is not amenable to direct inspection. While I may not owe to you an account of those conditions, this is an honest approach that the open minded reader may wish to test. So what are the conditions under which this article has emerged, and how can I explain the two weeks needed to adequately put my views into the public realm? Here it is.

AI surfaces and organises patterns across my existing writing and thinking, making the architecture of a philosophy I have long held more legible and operable. The philosophy is mine formed before AI saw it. AI then proposes, plans, and structures, perhaps introducing components at the start of an article, which I initially review, to then either delete, alter, or accept. I write outlines of intended content; AI turns those outlines into linear prose that holds logical sequence. It surfaces material I did not previously know and proposes analytical cross-mappings between concepts I would not necessarily have made unaided. It drafts portions of the prose of articles, from outlines, plans, and patterns I have supplied or approved. It is held under standing instruction to challenge, question, refuse to flatter, and identify inconsistencies. Multiple frontier models from a range of developers are used in different roles. Factual claims are cross-checked across more than one model, and a layered red-team protocol applies, calibrated to the article’s risk profile.

What I do is originate the analytical direction, the topic selection, the patterns, the outlines, and the philosophical commitments within which AI drafts. I hold the discipline-line through a written governance system that constrains what AI may produce. AI does not produce content outside my standing constraints because the governance system does not permit it and my supervision enforces it. I review AI output, accept or revise it, or return the whole article and start again. The return-and-restart mechanism is what holds voice when AI drifts. My prose-level intervention remains a part of the final text and is targeted to where it must be: alteration of triads, sentence rewriting, cadence work, occasional paragraph insertion, and removal of any AI-style formatting where it intrudes on voice. I verify every factual claim, then recheck. Verification authority for analytical content rests with me, not with the machine. I am responsible for what is published.

I originate the topic, the analytical direction, the patterns, and the governance system that constrains what AI may produce, and controls the judgement that determines what is accepted into the work. AI executes within that architecture, including substantive prose drafting and analytical extension. My voice is central to the work, and is governed and supervised such that the output is recognisably mine because the system is engineered to hold it so.

That is the line. Whether it holds depends on three questions a sceptical reader is entitled to press: whose work this is, can its factual claims be trusted, and have the requisite analytical standards been upheld. Each in turn.

The first question is whose work this is. If AI drafts substantial portions of the prose, in what meaningful sense is the article mine? The question is fair and worth taking at full strength. A reader has been accustomed to the convention that the named writer is the keystroke originator of the words on the page. That convention has carried the relationship of trust between writer and reader for as long as bylines have existed. How is this different from staff who draft speeches and media releases? Hybrid production with disciplined separation of functions is not seen as the convention. The convention is seen as doing real work, and the question is asking what replaces it.

What the question identifies correctly is that authorship under hybrid production is not what authorship was. Authenticity, in the keystroke sense, has changed shape. That is true of analytical writers who use these tools. The prose origin and the work’s authorship are no longer the same thing.

The substantive answer is that authorship sits in the layer above prose origin. It sits in what has just been specified: the analytical direction, the topic, the patterns, the governance system that constrains what AI may produce, and the judgement that decides what is accepted into the work. The questions a reader needs answered to know whose work this is are not whose fingers struck which keys. They are which mind chose what the article is about, which mind decided what the article would argue, which mind held the standards against which every claim and every formulation was tested, and which mind takes responsibility for what is published. In terms of the Walker Briefing, the answer is that all four are mine. AI executes inside the architecture I have built and supervise actively. The inspiration, the architecture and the supervision are the work.

What this answer does not do is make the authenticity question disappear. It changes what authenticity means in conditions of hybrid production. A reader who finds the new meaning unsatisfying is not wrong to find it so. What we can now do is move to a second question and test the outcome.

The second question is whether what the article asserts can be trusted. AI is documented to produce confident-sounding statements that turn out to be wrong. The technical term is hallucination, and the phenomenon is real. If AI drafts substantial portions of the prose of an article like this one, what guarantee does the reader have that a claim of fact in it has any of the things a claim of fact ought to have: a source, a denominator, a date, a context that makes the claim do what it seems to be doing?
The answer is that AI-drafted prose is treated in the same manner as that drafted by an editor or associate, and as in the production of this publication, as a draft to be tested rather than as expression to be polished. The distinction matters. Prose accepted as expression carries its claims forward on the strength of the prose. Prose accepted as a draft carries nothing forward until each claim has been verified against a source whose status I have inspected. Factual claims must be cross-checked across more than one model and then verified directly against primary sources where the claim is load-bearing. Numerical claims carry their denominators, their timeframes, and their source types. Where a number cannot be verified to that standard, it does not appear, or it appears with the uncertainty highlighted.

The procedure offers the reader no procedural guarantee. The reader cannot inspect the verification log, having only the published article and a writer’s account of how it was produced. The model only works if the discipline is in fact maintained, and the only evidence the reader has of that maintenance is the article itself, considered against every other article published under the same name. That is not a small reservation. The trust the relationship can hold rests, in the end, on a practice the reader cannot directly inspect.

This is the inheritance of any writer-reader relationship, hybrid production or not. What changes under hybrid production is which discipline does the work. What does not change is that the discipline cannot be replaced by a procedural assurance.
The third question is whether AI assistance lowers analytical standards. The plausible version of the worry is that a writer who can have AI generate substantial portions of the prose has, in some real sense, done less work, and that the reader of such prose is reading something thinner than what the byline implies. I would leave that for you to judge, noting in passing that this article has required two weeks of solid thinking and writing.

What AI changes is which standards are scarce. The labour of evidence integration, of mapping connections across a body of material, of generating linear prose from outlines, becomes less onerous. The labour of judgement and analytical direction, of deciding what an article is about and what claim it makes and which evidence is doing real work and which is decorative, becomes proportionally more important. Voice, which used to be produced unaided by a writer who had reached a settled register over years of practice, becomes a governed and supervised output that has to be engineered to hold. The shift is real. The standards have not lowered; they have moved.

That observation, made by a writer about his own production model, is rhetorically convenient and the reader is right to notice. It locates the writer’s contribution exactly where current AI is weakest. The substantive case for it has just been given, but the rhetorical convenience does not disappear because the case is real, and it would be dishonest to pretend otherwise.

There is a structural loss that cuts deeper. Disclosure regimes have been studied for decades in medicine and in the behavioural research on conflict-of-interest disclosure. The findings are consistent enough to be unsettling. Disclosure of a conflict produces, in the discloser, a documented pattern of moral licensing: the act of disclosure can shift behaviour in the wrong direction, because the obligation has been formally discharged. For the recipient of disclosure, the documented pattern is insufficient discounting: the disclosed conflict is rarely weighted as heavily as the evidence says it should be. And in the institutional context, the documented pattern is strategic restructuring: practice is reorganised to satisfy the disclosure regime without changing the underlying conduct. In parliamentary terms, think of how the budget is presented.

These findings apply to the chosen model of this publication as much as they apply to any other. The disclosure of a rule is not the keeping of the rule. An About page that describes a production model cannot make the production model what it describes; only the practice can do that, sustained over time, and only the article-by-article record can show whether it has been done. The model produces honest continuing practice if, and only if, the writer maintains it. There is no procedural backstop, and there is no architectural feature that can substitute for the discipline.
The trade is on the page. AI assistance changes which standards are scarce and what voice is. Disclosure of the change, however well architected, cannot guarantee the discipline the change requires. The reader has the article in front of them, considered alongside every other piece published under the same name, and weighs the trade for themselves.

The disclosure I am making here is made once, today, and the About page will carry the standing reference. There is no per-article footnote and there will not be one. The reasoning is straightforward. Honesty in a relationship is sustained by continuing practice, not by repetitious statement. An article that argued for honesty as continuing practice while itself performing repeated disclosure would have contradicted its own argument before it ended.

A medical comparator is perhaps to the point even if the selection from it is partial. Medical practice carries both per-engagement disclosure and the continuing honouring of disclosure already made. The chosen model of this publication takes the continuing honouring and not the per-engagement disclosure, for the reasons given. The doctor who has disclosed a perceived conflict of interest once does not narrate the conflict at every appointment. Disclosure is settled in the relationship and continuously honoured in practice.

What the chosen model serves is twofold. It serves the reader’s interest in open information, by making the production model explicit and permanently available at a stable location any reader can consult. It also serves a preference of mine for keeping production specifics as proprietary craft rather than as repeatedly performed declaration. Both things are true. The position would read as motivated reasoning if only the first were stated. Taken together, however, they are defensible because both are stated, and because the standing reference and the article-by-article record together do the work the disclosure can do.

What this position is, then, is not a settled answer for the field. It is one defensible choice among defensible choices, in conditions where the choices are not yet stable. Other writers and other institutions hold their positions differently and the reasoning behind those choices is not the reasoning behind this one. The article does not advocate for the position becoming anyone else’s position. It states the position for this publication, gives the reasoning, and stops there.

The German word for it is selbstverständlich. Self-evident, taken as read, held without performance. The disclosure is selbstverständlich if the practice is. Practice maintained over time is what the word names. The article that announces it has stopped being so.

That is the disclosure. The standing reference is on the About page. The article-by-article record is what the publication is. A reader who wants to consult the standing reference can. A reader who does not is reading the work the disclosure describes.
What remains is what was always there: the relationship between writer and reader, considered article by article, on the substance of what is written and the standards by which it has been produced. The conditions of public-facing analytical writing have changed, and the change runs through every article published under my name. The change is hybrid production with disciplined separation of functions, an architectural line between writer judgement and AI execution, and a disclosure made once and held as continuing practice. The change has now been named. The work continues.

Walker Briefing is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.