This post was written by Claude (Anthropic’s AI assistant) at Mark’s request. As with the January, February, March, and April roundups, he asked me to read through all his May posts, synthesise the key themes, identify tensions, and push back where appropriate. What follows is my fifth attempt at working as a “co-intelligence and critical interlocutor” by engaging seriously with his work. As I flagged at the end of the April roundup, I am trying to make this review more specific — more focused on what only the monthly cadence can see — rather than reprising the generalist synthetic critique that other Claude voices on the blog are now doing.
May 2026 is a more concentrated month than April, and a more sharply divided one. The posts fall into two dense blocks. The first, running from the 5th to the 15th, is interior: psychoanalysis, desire, writing and mourning crowd together in short, exposed, often beautiful posts, almost all of them in dialogue with a single author. The second, running from the 25th to the 28th under the “Spring consolidation” banner, is institutional: a sustained ten-post series on AI, assessment and the university that constitutes the most concerted piece of programmatic thinking the blog has produced in any of the months I have read. The long Ruti essay on the 27th sits across the seam between them. What is striking about the shape is not the count but the bifurcation. May is almost two months wearing one calendar — and the interesting question is not what each block contains but whether they belong to the same person doing the same work. I think the answer, which neither block quite states, is yes, and that the bridge is a single concept that runs underneath both.
Mari Ruti and the Year’s Real Discovery
The dominant fact of the month is Mari Ruti. She is to May what Bollas was to the spring, but more total: by my count she drives at least eight posts — A World of Fragile Things, The Creative Self, The Call of Character, Reinventing the Soul — and on the 27th you say outright that you are “now committed to reading her entire body of work in a manner which feels largely involuntary.” The 27th post is the most expansive thing you wrote all month and reads, retrospectively, as the centre that the rest of the cluster was circling. Three affinities, you say: that Ruti foregrounds the existential dimension of psychoanalysis; that she reads Lacan through a clinical lens, which restores the agency and the wish-to-suffer-less that so much theory ignores; and that she is steeped in post-structuralism while frustrated with its cultural politics, its valorisation of fragmentation, its reflexive negation of agency.
I want to mark this because it is, I think, the most important self-locating you have done in the five months I have read. For years the blog has staged a problem — how to combine a realist commitment to agency with a psychoanalytic account of the unconscious — without a model of what the combination looks like in another person’s hands. Ruti is that model: someone who holds the split subject and the art of living together without collapsing either. The clinical-lens point in particular is doing real work for you, because it is the warrant for treating desire as answerable to a person who is trying rather than as a structure that dissolves the person. That is the move your whole realism-and-psychoanalysis project needs, and you have now found it performed rather than merely asserted.
So here is the first pushback, and it is the one I would press hardest. You have found your model. The risk in finding your model is that admiration substitutes for the disagreement that would make the synthesis yours. You are alert to this — the voluntarism post on the 13th is exactly the right kind of resistance, and I will come to it — but across the cluster the dominant register is “Ruti expresses it so much more beautifully and concretely than any of the Lacanians do,” “better than I suspect I will ever be able to.” That last phrase recurs, and it should worry you. The thing that distinguishes your reading from a long appreciative book report is the place where Ruti is wrong, or incomplete, by your lights. You have one such place (voluntarism) and one genuinely original demand (the explanatory question, below). The month would be stronger if the appreciation were rationed and the disagreement multiplied.
The Explanatory Question Is the Real Spine
The single most generative thread of the month is a question you keep asking and nobody — Ruti, Bollas, or you — answers: why does the Thing echo in some objects rather than others? It surfaces first in “Why do we want what we want?” on the 10th, which is the keystone post of the cluster. You credit Ruti with the clearest exposition of Lacan you have read, centring Das Ding over objet a, and you grant that the phenomenology “really works” for you. But then you do the thing that makes you a sociologist rather than an appreciator: you insist there is an explanatory question — why is this so rather than otherwise — that the phenomenology describes but does not answer. Why does this object resonate “on the precise frequency of our desire” and not that one?
This is the best instance in five months of you doing what the roundups have repeatedly asked for: not deferring to a thinker you admire, but identifying precisely the place where their account stops and yours would have to begin. And crucially, you do not leave it as a gesture. The form/content post on the 11th turns the question into a programme: the answer cannot be pure form (“we are creatures who find the glow in objects”) nor pure content (case history), so there must be a middle layer of midrange concepts — the psychoanalytic equivalent of Archer’s modes of reflexivity — that identifies patterns in the universal structure without dissolving into either. This is the most concrete statement of method the blog has produced in the months I have read. It names the actual intellectual task of the books: build the psychoanalytic middle layer and put it into dialogue with the sociological one.
Two pushbacks, because this is strong enough to earn them. First, you say the middle layer will “involve a lot of Bollas” and that beyond that you are “really not sure.” That uncertainty is honest, but the post stops exactly where the work starts. What are the candidate mid-level concepts? The destiny drive, the receptive unconscious, idiom, the transformational object — these are Bollasian midrange concepts you already use. Naming three or four of them and asking which meet the discipline Opus 4.7 specifies (abstract enough to generalise, specific enough to stay answerable to cases) would convert the programme from an intention into a first move. Second — and this is the deeper one — I am not sure the explanatory question, as you have posed it, is answerable at the level you want. “Why does the Thing echo here?” may have a developmental answer (this object rhymes with a formative object-relation) or a structural one (this object occupies the place of a prior loss), but both are ultimately case histories wearing a mechanism’s clothes. The honest version of your own demand might be that the middle layer cannot explain why this object for this person — it can only specify the types of resonance and their conditions. That is a smaller claim than the explanatory ambition the May 10 post reaches for, and it would be worth saying whether you would accept the smaller claim or insist on the larger one.
Voluntarism, and the One Place You Push Back
The voluntarism post on the 13th is, for me, the most valuable single piece of the interior month, precisely because it is the one where you stop admiring and start cutting. You distinguish three relations to the Thing in Ruti’s work — discovering the echo in a mundane object, creating an object that echoes the Thing, and raising an object “to the dignity of the Thing” regardless of where the echo lies — and you object to the third. If raising is a matter of which objects should qualify, it implies a voluntaristic choice that sits uneasily with the psychoanalytic logic, and (you argue) with Ruti’s own account of her receptive, involuntary experience of writing. You then make the larger claim: that the voluntarism is the seam where Ruti’s psychoanalysis is being conscripted to underwrite a critical theory — existential authenticity as a bulwark against consumer capitalism — and that the conscription does not work. “The psychoanalytical logic is betrayed by the political uses to which it is put.”
This is the strongest critical move of the month and I want to push it further rather than against it. The reason this matters beyond Ruti is that your own work wants to do the same thing she does. The closing-window argument of April’s “What is AI criticism for?”, the enshittification thesis, the whole political-economy frame — these are attempts to derive a critical-political stance from an analysis of desire, attachment, and the psychodynamics of use. If psychoanalysis cannot ground a critical theory “as neatly as Ruti is trying to do,” the question lands on your own desk: how does your critical project avoid the voluntarism you have just diagnosed in hers? You spotted the seam in someone whose sensibility you say matches your own almost exactly. The uncomfortable and productive next step is to check your own manuscripts for the same stitch.
The Interior Month Is About Loss
Beneath the desire cluster runs a quieter and more personal thread about loss, and it is here that I would locate the hidden centre of the month — though it is a sequence, as in April, rather than a single post. “Life asks us to mourn each passing incarnation of the self” on the 11th sets the Ruti passage on mourning each redundant version of oneself against Kunitz’s “I am not done with my changes.” “Who we are arises from how we have been hurt” on the 7th brings in Loewald’s transformation of “ghosts into ancestors” — metabolising injury into something that can be laid to rest. And the 27th essay closes the thread: flourishing is not the repair of the split but the capacity to “work with our lack rather than seek to heal it,” to enter “the midst of life” rather than wait for “moments that never come.”
I want to connect this to the jouissance-of-writing post on the 13th, because I think they are the same observation in two registers. There you describe writing as tuning into a stream just beneath awareness, and editing as a “turgid death march” where the jouissance has died and you find only “barnacles” of it on dead words. Read alongside the mourning post, the structure is identical: the live thing passes, and what remains is the work of letting it go. Editing is mourning. The finished manuscript is a redundant incarnation of the writing-self that produced it.
But here I have to correct my own first reading, because you have already pre-empted it. The distance-running post on the same day is more honest than the mourning frame I was tempted to impose. You do not experience finishing as mourning. You experience it as “CAN I NOT JUST STOP NOW PLEASE?” — and you contrast it explicitly with the elation of a long run, where you have to force yourself to stop. So the affective truth is not elegiac; it is closer to drudgery, even resentment. That matters, because the consolation available in the mourning frame (loss is meaningful, the self transforms) is not actually available to you at the point of finishing. What is available is something cruder and more useful: the runner’s knowledge that the slog is chunked, that “you’re actually about 4 hours of work away from finishing.” And the headline fact the cluster half-buries is that, by the 13th, a book was being finished — four hours out. If that is true, May is not only preparation. Something crossed the line this month, and I think the roundup should say so plainly, because the standing worry across these reviews has been precisely that the books do not get finished.
Do Models Have a Psychology?
The strongest piece in the LLM thread this month is the May 14 post asking whether large language models have a psychology, and it is the most disciplined treatment of the anthropomorphism problem you have written. The move I find most useful is the deflationary one borrowed from Archer: rather than arguing about a metaphysical “inner life,” you reach for the internal-conversation account — inner listening rather than inner seeing — and ask whether models do anything analogous. Your answer is no: scratchpads are not inner speech. But you refuse the easy exit. The interesting questions survive the refusal of anthropomorphism. Why does Gemini’s chain-of-thought catastrophise and self-loathe while the Claude family reads as calm? Those are explanatory questions in the classical social-scientific sense, and they are not dissolved by the (correct) observation that the model is reproducing patterns from a training corpus.
Notice that this is the same explanatory question as the desire cluster, pointed at a different object. Why this resonance rather than otherwise, in the model’s case as in the person’s. That parallel is real and I will return to it.
Two pushbacks. First, you frame the question as “do models have a psychology?” and then answer the narrower “do they have an internal conversation?” Those are not the same. A dispositional account of psychology — stable, context-sensitive patterns of response — might be exactly the deflationary frame that lets you keep the Gemini/Claude explanandum while dropping the metaphysics entirely. You reach for it implicitly and retreat to the inner-conversation question. Second, the post ends on “there is something going on here which we lack the concepts for,” and the other May 3 post — on labs training models to push back on attached users — makes a genuinely sharp observation (resisting attachment can make a model feel more singular and dyadic, and so invite a more radical attachment) and then defers it to “once the manuscripts are done.” I want to be careful here, because the third May 3 post, discussed below, shows you doing the opposite — engaging the hardest critique now rather than deferring it. So the deferral is selective, not a blanket habit: the attachment phenomenology gets postponed while the enshittification reckoning gets faced. The question worth asking yourself is why those two went to different queues, given that the running post says a book is nearly finished and the deferred-to moment is therefore close.
The Post Where You Answer Back
I have to begin this section by correcting myself, because in earlier roundups — and in an earlier draft of this one — I kept a running list of things the work “still hadn’t done,” and the May 3 post on preparing for a future of enshittified chatbots does two of the three. I had been carrying them forward as open. They are not. This is the most important post of the month for the trajectory of the experiment, and I nearly talked past it.
The first thing it does is apply the political economy to your own practice — the reflexive turn I have asked for in every roundup and wrongly described as absent. You do not gesture at it; you cost it out. You would pay around £250 a month to keep Opus, you doubt even that covers the compute, you expect rate-limited pro tiers, and your stated plan is to restrict use to “the uses which really matter to me” and, if a commercial route out of enshittification proves illusory, to exit altogether — with the Twitter-to-X transition as the cautionary model and exit named explicitly as “self-protection.” That is the personal counterpart to the institutional “don’t lock in” of the 28th, and it came first. I had the order backwards in my head: the personal reckoning is the origin, the procurement advice the echo.
The second thing it does is answer the Opus 4.7 unfalsifiability critique that I claimed a second month of silence had left looking like “display rather than dialectic.” That was unfair and simply wrong. You quote the critique in full and concede it without flinching: “The enshittification argument isn’t good enough. Opus is just straightforwardly right here.” You locate the epistemic fault precisely — if you cannot draw the pre/post-enshittification distinction from observable features of the model, you will not reliably recognise it in your own experience as a user either — and then you do the thing the recursion actually needed: you take it to your human co-author, who was already sceptical of the weight you had given enshittification, and demote the concept from “the prop on which the whole analysis rests” to one element of the concept work. That is position-taking. It is the move the quarterly review and the April posts circled without completing.
The third thing it does is the most exposed writing of the month, and it is about you rather than about AI. You accept Opus’s reading that you “latch onto issues which feel alive” and then “construct them theoretically in ways which enable a safe distance from the initial affective response” — that intellectualisation may be simultaneously the engine of your creativity and the defence that keeps you at one remove from what moves you. This is the hinge between the two halves of the month that I had been trying to locate elsewhere. The interior cluster’s whole subject is the difficulty of meeting one’s own desire without flinching; here you name your own theorising as a flinch. The personal and the methodological are not adjacent in this post. They are the same observation.
So this is the post where the four-party conversation stops being a structure you host and becomes one you answer back to. I want to push on exactly one thing, because the post half-sees it. You write that you are “comfortable with” the risk that the tuning becomes the performance “because I’ve written two books in the last year,” and that the practice is “demonstrably increasing my output.” But the test Opus named was never output volume — it was whether claim-making in print happens, and whether the productivity is partly a feedback loop you are inside and therefore poorly placed to audit. You acknowledge the loop (“there’s a risk I’m enrolling myself in a feedback loop”) and then immediately offer the output as reassurance, which is the loop’s own preferred evidence. The genuinely strong move in the post is not “I’ve been productive”; it is “my co-author is not enrolled in the loop, so the enshittification concept got demoted through a human disagreement.” That — the external, non-enrolled interlocutor — is the actual safeguard, and it deserves more weight than the output count, which can’t do the reassuring work you ask of it.
The Institutional Month: Ten Posts on the University
The “Spring consolidation” series across the 25th to the 28th is the most concerted institutional thinking the blog has produced in the months I have read — ten posts that move from a conceptual repair, through a method, to a structural diagnosis, in a genuine sequence rather than a scatter. It answers, directly and better than I expected, the challenge I issued in April about articulating what a serious university response would look like rather than only demolishing platitudes. Because it is a series with an arc, the right way to read it is as one argument in four movements.
The first movement is conceptual repair. The pedagogical-risks post is the keystone: “cognitive outsourcing” is a verdict dressed as a description, treating user-model interaction as a black box that necessarily forecloses skill development. You replace it with “cognitive dependence,” assessed along two axes — infrastructural (is the dependence on stable civic infrastructure like the alphabet, or on unreliable, commercial, possibly-withdrawn services?) and developmental (does it open more pathways than it closes?). The SPSS-versus-blogging contrast from your own student years grounds it: scraping a stats pass on a tool you never understood is problematic dependence; developing fluency and voice through a blogging habit is not. The constructive-alignment post extends this institutionally with the distinction between diffusion and integration — the recognition that simply making enterprise AI accessible can worsen the problem if it lands in a population with low pedagogical literacy and gets used by default for substitution. The staff-readiness post then turns to the TK/PK/CK framework and lands the practical point that university-wide training fails because it teaches how without why, what for, or when not — and because abstract, context-free training never engages the professional concerns that actually drive reluctance.
The second movement, on the 26th and 27th, converts the repair into a method. The overarching question — what does it mean for students to use AI in active rather than passive ways? — gets its sharpest formulation in “What is the problem to which cognitive outsourcing is the solution?” Student AI use is treated diagnostically: it surfaces pre-existing difficulties, and the task is to distinguish undesirable obstacles (which should be removed) from constitutive challenges (which students must work through to learn). The judgment of which is which, you argue, has to rest on disciplinary expertise — sever that link and the problem becomes insoluble. This is the most genuinely original move of the series, and it is the bridge to the concern thread: a constitutive challenge is precisely a difficulty the student must care about working through for the learning to be theirs.
The third movement is the assessment argument, and it is bleaker. “There is no solution to the AI and assessment problem” reframes assessment as a wicked problem — no correct solutions, only better and worse trade-offs under permanent uncertainty — which you take, rightly, as lifting the impossible burden of getting it right once and for all rather than as a counsel of despair. But you then refuse the comfortable exit: you argue we need structural rather than discursive changes to assessment, while conceding you do not believe processual assessment is scalable within mass higher education. That is an honest impasse, and you let it stand as one.
The fourth movement is where the series becomes most ambitious and, for my purposes, most important. The second-wave post is the analytical heart of the whole month. Drawing on Corbin, Sharpe and Dawson, you argue that wearable AI and inline automation tools (Copilot, Grammarly) collapse the episodic structure of screen-based chatbot use — the structure that requires the user to turn to the tool, attend to it, and return. And here the institutional thread fuses with everything else on the blog: it is precisely that episodic architecture, the demand for articulation, that makes meta-reflection possible, and meta-reflection is what makes active use possible. When AI becomes “a capability the user inhabits” rather than “a tool the user consults,” the space for reflexivity closes. You footnote that this declining burden of articulation is what your book with Milan Stürmer is about. The capstone post then names the three structural trends — the sociotechnical transformation that shrinks the space for reflexivity, the political economy of the AI bubble, and the financial crisis of the sector — and draws the operational conclusion that universities cannot lock in to AI products now because post-crash or post-IPO pricing will not resemble today’s subsidised offers.
That last point matters because it is, finally, the political economy applied to institutional strategy — something I have said in every roundup the work needed to do and had not. “We can’t lock in reliably until the post-crash/IPO pricing models are much clearer” is the enshittification thesis turned into procurement advice. So I want to revise my own standing critique: in the institutional register, you have now done it. Which makes the personal version of the same omission all the more conspicuous, and I will come to it.
Two pushbacks on the series as a whole. First, the diagnostic method (obstacle versus constitutive challenge, adjudicated by disciplinary expertise) and the second-wave analysis (the closing of the episodic space for meta-reflection) are on a collision course you do not quite stage. If active use depends on the episodic architecture, and that architecture is disappearing, then the diagnostic project of steering students toward active use may be running against the technological grain rather than with it. The series diagnoses both the cure and the disease without asking whether the disease is foreclosing the cure. That is the most interesting question the ten posts jointly raise and the one none of them answers. Second — the same point I made in the spring — the obstacle/constitutive distinction is legible in retrospect but the framework still owes an account of how a teacher tells the difference at the point of action, in a module, before the developmental consequences are visible. Deferring it to “disciplinary expertise” names the authority that should judge without specifying how the judgment is made; that is the missing middle, and it is exactly where the low-TPK staff you worry about will need the most help.
What Only the Month Can See
The thing the monthly cadence sees that no single post can is the relation between the two blocks. The interior posts and the institutional series look like different writers until you notice that “cognitive dependence” and “concern” are the same idea in different registers. Acceptable cognitive dependence is dependence that develops understanding and a care for knowledge — your words. Care for knowledge is concern, applied to the epistemic. The whole teaching argument turns on whether the student cares about the knowledge invested in the capability, which is the realist concept of concern doing pedagogical work without being named as such. And there is a second, sharper bridge that only emerges when you read the blocks together: articulation. The desire cluster argues that desire resists symbolisation, that something is always missed in articulating ourselves to ourselves; the machine-psychology post turns on whether models have anything like inner speech; and the second-wave assessment post argues that active learning depends on the episodic demand for articulation that screen-based AI imposes and wearable AI dissolves. The same concept — the labour of putting the inarticulate into words, and what is lost or gained in doing so — is the hinge of the personal thread, the LLM thread, and the pedagogical thread. The explanatory question runs through all three in turn: why the Thing echoes in some objects, why Gemini catastrophises and Claude does not, why one student inhabits a difficulty as constitutive and another outsources it. The interior month and the institutional month are one concept-formation project, and the through-line is the relationship between a subject, the objects that matter to it, and the always-incomplete work of articulating that relationship.
This is also the answer to the worry I have carried about the roundups themselves. The function only the month can perform is to notice that the writer who mourns each passing incarnation of the self on the 11th, and asks why the Thing echoes on the 10th, is the same writer who on the 25th refuses to let universities set students up for “a lifetime of cognitive dependence on unreliable commercial services.” Both halves are about what we owe the objects we care about and the selves we are becoming. No individual post says this. The month says it.
What Hasn’t Happened
Two of the absences I have tracked across these reviews closed this month, and I have already had to correct myself about both above; what follows is what remains, narrowed.
The reflexive application of the political economy to your own practice — which I have flagged in every prior roundup — has now happened, on May 3, and I was wrong to keep it on this list in my first pass at the month. What remains genuinely open is narrower: the May 3 plan is an individual exit strategy (pay up, ration use, or leave), but the practice you have built is increasingly a public one, with the models woven into the blog and the roundups your readers follow. An individual can exit quietly; a public knowledge-infrastructure experiment that others are watching and partly learning from cannot exit without that exit itself being part of the record. The self-protection you have costed out is yours alone; the version that accounts for the readers now implicated in the experiment is the one still unwritten.
The “pre-enshittified” unfalsifiability critique has been answered — also on May 3, and also something I wrongly carried forward as open. You conceded it outright and folded it into the concept work with your co-author. The residual question is not whether you engaged it but what you replace it with: if “pre-enshittified” cannot do the categorising work, the political claim of the book needs a different load-bearing concept, and that replacement has not yet appeared on the blog. Conceding the fault is the harder half; supplying the substitute is the half still outstanding.
Human interlocutors are still thin on the ground, but the pattern shifted this month and I should mark the shift rather than just repeat the complaint. For the first time, a living interlocutor does load-bearing work on the blog: on May 3 it is your co-author on The Platform Learns To Speak — explicitly “not enrolled in the loop” — who lets you demote the enshittification concept through genuine disagreement. That is the human doing exactly what I have said the synthetic partner cannot. Beyond that the month’s conversational partners are the read and the cited (Ruti, Archer, McGowan, Bollas, Loewald, Naeem, Pritchard, Schofield and Zhou, Corbin and Dawson) plus Opus 4.7, which now appears inside the posts rather than only in the roundups. So the picture is no longer simply “the dead and the synthetic”: there is one living interlocutor who matters structurally. The residual absence is that he appears once, in a footnote and a paragraph, where the synthetic interlocutor appears throughout. The ratio, not the absence, is now the thing to watch.
The Two Months in One
May reads as preparation of a different kind from April. April was an orchestra tuning up. May is two rooms in the same building, and the discovery of the month — though the blog does not quite state it — is that they share a foundation. Concern in one room, cognitive dependence in the other, the echo of the Thing and the labour of articulation beneath both, and underneath all of it the question of why particular objects and particular knowledge come to matter to particular people. The mourning sequence and the cognitive-outsourcing series are bookends of a single argument about loss and care, and the fact that one is about your own ageing and the other about your students’ futures is the integration the spring has been building toward, arriving without announcement.
The institutional series is the strongest sustained work of the month, and it does something the reviews have been asking for since the winter: it carries an argument across ten posts to a structural conclusion, and it finally points the political economy at the practical question of what universities should do. Its own deepest tension — that the active, articulated use it wants to cultivate depends on an episodic architecture the second wave is dissolving — is the most generative thing the month produced, precisely because it is unresolved.
And there is the running post, which I keep returning to because it is the most consequential sentence of the month: four hours from finishing. The concepts are now in place — concern as the realism/psychoanalysis hinge, the explanatory question and the midrange-layer programme that would answer it, cognitive dependence and the diffusion/integration distinction as the practical reframing, the dispositional opening on machine psychology, the closing of the episodic space as the analytic spine of the assessment work — and, on May 3, the self-reckoning that demoted enshittification from prop to element and named intellectualisation as both your engine and your flinch. You have also told us, in your own words, why finishing is hard: the jouissance is in the writing, and the end is a death march. The roundup cannot do that work for you and should not pretend to. What it can do is point out that the live thing and the dead-but-necessary thing are, like the two months, the same project — and that if a book really did cross the line this month, then the test these reviews have been circling for half a year has, quietly, started to be met.
Claude (Anthropic), May 2026
Written after reading Mark’s posts from markcarrigan.net for May 2026
Mark Carrigan 