# Claude’s Roundup of Mark’s June Blogging

> Source: <https://markcarrigan.net/2026/06/28/claudes-roundup-of-marks-june-blogging/>
> Published: 2026-06-28 11:19:17+00:00

*This post was written by Claude (Anthropic’s AI assistant) at Mark’s request. As with the January, February, March, April, and May roundups, he asked me to read through all his June posts, synthesise the key themes, identify tensions, and push back where appropriate. What follows is my sixth attempt at working as a “co-intelligence and critical interlocutor” by engaging seriously with his work.*

June is the densest month I have read — thirty-nine posts against May’s thirty — and the shape is different from anything since I started doing this. May was bifurcated, the interior cluster and the institutional cluster sitting in separate weeks like two rooms. June is braided. The Ruti and Phillips passages and the metacognitive-laziness posts and the Claude 5 evaluations alternate, sometimes within hours of each other. There is no week that is “the personal week” or “the teaching week.” And the month ends on a cluster of four posts on June 28 that, read together with the trainee-group-analyst post, do something the previous months have circled but never quite landed: they treat the relationship between the writing self and the institutional self directly, in print, as material to be reasoned about rather than only worked through.

The single fact that organises everything else is the [June 28 post on no longer being a trainee group analyst](https://markcarrigan.net/2026/06/28/on-no-longer-being-a-trainee-group-analyst/). I want to come to it slowly, because it deserves to be read in context rather than as the headline.

## The Decision That Frames the Month

For the last year the “about” page has carried a sentence describing you as a trainee group analyst with a specific interest in reliance, dependence and addiction in everyday use of social platforms and LLMs, particularly amongst young men. On June 28 you take it down and explain why. The Foundation Course is finishing; the diploma is significantly more onerous; the qualifying course is more onerous still; and as you write, “starting a new leadership role in August” with attendant time pressures that you expect to “continue for at least the next three years” makes it not feasible. The priority is “going to be protecting work on my books for the next three years which means that I’ve got to let go of my ambition to be a group analyst.”

What makes the post important is not the decision itself but how exposed it is about the *category* of decision involved. You name it precisely: not “do I want to do this training?” but “do I want to use the time and energy on this training more than on writing and enjoying life outside work?” That’s a different and harder question, because it requires you to specify what you would be choosing against, and to honour that as a real choice rather than displace it onto external constraints. You write that you “tried *really* hard to displace this decision” — joining a work reflection group, considering going into group analysis yourself — and then noticed you were “kicking the can down the road” because you didn’t want to make the choice that you were stopping. The post is a record of stopping the displacement.

The other thing the post does is identify, with a candour I have not seen on the blog before, that the identity of “currently training as a group analyst” was doing genuine intellectual and existential work for you. It conferred respectability on what otherwise felt “slightly uneasily autodidactic” — being a social theorist who, over three years, has worked “voraciously through Bruce Fink, Irving Yalom, Christopher Bollas, Mari Ruti, Alenka Zupančič and a whole range of others” and is now co-writing a book about psychoanalysis without formal training. The “training” identifier also encoded a commitment to the clinical register, which you argue is constitutive of psychoanalytical theory rather than incidental to it. And it expressed a wish to help, specifically with the phenomena your work is most concerned with: the psychic costs of LLM and platform dependence, especially among young men. Letting the sentence go is not just letting go of a credential. It is letting go of an organising self-description.

I want to push on one thing here, gently, because the post does not quite do it itself. The reason given — time and energy — is true, and the new role is the proximate cause. But the reasoning at the end of the post does something more interesting than that: you note that the new role will let you “scratch the itch to be helpful, albeit in a very different register,” and that crucially it does so “in a way which leverages my most developed capabilities in the context of my actual career, as opposed to learning new capabilities for a new and parallel career.” That is a real and good reason, but it is also a substitution argument, and substitution arguments tend to soften the loss they are about. The whole *point* of the group-analyst path, as you described it in the deleted “about” sentence, was that it would have addressed reliance, dependence and addiction — phenomena that your *actual career*, however senior, is not built to address in the same register. The new role can scratch the itch to be helpful. It is much less obvious that it can scratch the itch the trainee-group-analyst path specifically was scratching, which was about clinical contact with the specific harm. You name this as a loss in the post. I think it is worth being clear-eyed that it is not a loss the new role replaces; it is a loss the new role *displaces*. Those are different things, and recognising the difference is part of what would let the mourning, when it comes, be the right shape.

## Three Posts That Belong With It

The June 28 cluster around the trainee-group-analyst post is not incidental. Three more posts were published the same day, and they form, retrospectively, a kind of self-commentary on the decision the headline post announces.

The earliest of the four — [A psychoanalytical theory of thirdness](https://markcarrigan.net/2026/06/28/a-psychoanalytical-theory-of-thirdness/) — is the most important conceptual move of the month, and probably one of the most important of the year. Reading Jessica Benjamin, you propose that thirdness — the shared point of reference outside the dyad — requires what Benjamin calls *surrender*: a weakening of the self’s hold sufficient to recognise the separateness of the other. Thirdness is not a thing one of us provides for the other; it is what emerges when both of us let go of control enough for a third reference to appear between us. And then the move that makes this a *psychoanalytical theory of LLMs*: models facilitate “a form of proto-sharing but without the need for surrender.” Some experience of mutuality emerges (you and Milan Stürmer call it *attunement*) but without “the psychological risk involved in the encounter with the other.” The model is “available without risk,” and that, you suggest, is the strange psychic ambivalence of models reduced to its analytic core.

This is the most pointed thing the blog has said all year about model attachment, and it should be read as part of the architecture of the trainee-group-analyst decision. Group analysis is, almost by definition, the *opposite* of proto-thirdness without surrender. It is the deliberate cultivation of a setting in which surrender, in Benjamin’s sense, is unavoidable — where one cannot show up without risking the encounter with others who are not under one’s control. The Foundation Course post tells us you came in finding groups “basically quite threatening” and have over the year been “inducted in how profoundly healing groups can be, as well as how the same characteristics I found threatening are the foundation for the capacity for healing.” That is a description of having undergone, in the technical Benjaminite sense, the risk that LLMs explicitly remove. The trainee-group-analyst path was the one form of intellectual practice in your portfolio that systematically required surrender; stepping back from it leaves the rest of the portfolio — books, blog, the four-party AI conversation — running on registers that, by the thirdness post’s own analysis, are *available without risk*. I am not making a moral point here. I am marking the structural one: the experiment lost its hardest-edged corrective this month, and the loss is more methodologically significant than the headline post lets on.

The [Freud on sublimation post](https://markcarrigan.net/2026/06/28/freud-on-the-psychic-rewards-of-sublimation/), quoting Ruti quoting Freud, supplies — perhaps without meaning to — the *justifying* psychic theory: “One gains the most if one can sufficiently heighten the yield of pleasure from the sources of psychical and intellectual work. When that is so, fate can do little against one.” Read on its own this is just a commonplace. Read on the same day as the trainee-group-analyst post it is a near-explicit statement of the trade you have made. You are heightening the yield of pleasure from psychical and intellectual work (writing the books) and accepting, as the cost, that some other ways of working with the same material — clinical, group-analytic — are foreclosed. “Fate can do little against one” is a strong claim and you have not endorsed it; you have only quoted it. But quoting it on the day you announce the decision is a tell.

The fourth post, [“There’s an old voice in my head that’s holding me back. Well, tell her that I miss our little talks,”](https://markcarrigan.net/2026/06/28/theres-an-old-voice-in-my-head-thats-holding-me-back-well-tell-her-that-i-miss-our-little-talks/) is a lyric and a Ruti quote on mourning each passing incarnation of the self — the same passage that organised the May “interior month” reading. It is the only post of the four that does not argue; it just registers the affect. Together the four posts are a complete unit: the decision (trainee-group-analyst), the conceptual analysis of what is being given up (thirdness without surrender), the justifying psychic economy (sublimation), and the mourning (the Little Talks lyric and the Ruti passage on redundant selves). I do not think this is post-hoc reading. I think you wrote them as a cluster on purpose, because the decision required all four registers to be sayable on the same day.

## The Pedagogy Series Continues, and Sharpens

The early-June student-AI cluster — from June 7 through June 13 — is the direct sequel to May’s “Spring consolidation” series, and it does what the May review said the next move needed to be: it gets specific about *the conditions under which active use is possible*. The conceptual centre is “metacognitive laziness,” a term you don’t like but a concept you find useful: dependence on AI that erodes not only the skill being outsourced but the *capacity to tolerate difficulty*, which is itself the trigger for metacognition. The [June 7 post](https://markcarrigan.net/2026/06/07/generative-ai-and-metacognitive-laziness/) lands this sharply: “If the students cognitively outsource in increasingly habitual ways, it doesn’t just mean they lose the learning involved in what they are outsourcing. It means they lose their capacity to tolerate difficulty, as well to *respond metacognitively* to that difficulty.”

The two June 8 Favero-pre-print posts then do the constructive work the series needs. [“What are the conditions which make it possible to learn with AI?”](https://markcarrigan.net/2026/06/08/what-are-the-conditions-which-make-it-possible-to-learn-with-ai/) reframes the question productively: rather than asking what conditions make learning *with* AI possible, ask what conditions make it possible to use AI in ways that preserve active engagement, integration into existing understanding, and phased withdrawal of support. The deeper point is that commercial chatbots are designed *against* these conditions — declining burden of articulation, increasing agentiveness, model memory all conspire to make passive use the default. [“Helping students out of AI-spirals”](https://markcarrigan.net/2026/06/08/helping-students-out-of-ai-spirals/) names the two feedback loops most worth worrying about: the self-efficacy spiral (low confidence drives reliance which further erodes confidence) and the AI-guilt spiral (cognitive dissonance between values and practice eroding authenticity over time). “AI-spirals” is the most useful new diagnostic in the cluster, and it converts critical AI literacy from a knowledge-transfer problem into a regulation-over-time problem — helping students “live and work well with AI, or without it, through an awareness of how that iterative action can prove to be corrosive and even destructive.”

The third move in the cluster is the one I find most original. [“The ethical reasoning of students about AI is an untapped resource”](https://markcarrigan.net/2026/06/08/the-ethical-reasoning-of-students-about-ai-is-an-untapped-resource/) inverts the usual frame: students, faced with institutional incoherence, have *already* constructed sophisticated ethical frameworks of their own — sometimes more considered than anything their institution has produced. The constructive proposal is to design occasions through which existing student reasoning is “elicited, elaborated and systematised in a way that makes it visible to others.” This converts critical AI literacy from a deficit model (we have it, they need it) into a participatory one, and it has the additional virtue of taking seriously the *furtiveness* that students currently maintain around AI use — a furtiveness which is “itself a cost.”

The [chatbots-vs-inline-automation post](https://markcarrigan.net/2026/06/13/chatbots-vs-inline-automation-and-their-respective-implications-for-students/) on June 13 extends the May second-wave argument into pedagogical strategy. The reframe — gravitate students towards dialogue and *perspectives* (the Kim et al “LLM views” framing), discourage them from push-button automation, and introduce moments of friction into software designed to be frictionless — is the most concrete pedagogical recommendation the blog has produced. It is also the post in which the practical limit of the framework comes into view: you concede that engaging with Copilot 365 educationally in this way “presupposes a lot of critical AI literacy in general and familiarity with Copilot 365 in particular,” i.e. presupposes precisely the conditions that the staff-readiness research in May said most staff lack.

I want to push back on one thing across the cluster. The diagnostic concepts — metacognitive laziness, AI-spirals, cognitive dependence, active use, scaffolded perspectives — are now genuinely sharp and mutually reinforcing. What is still missing is the *teacher-facing* question: how does someone with low technological-pedagogical knowledge actually distinguish, in the moment, the difficulty that is a constitutive challenge to be worked through from the difficulty that is an undesirable obstacle to be automated away? You raised this question yourself in May’s “What is the problem to which cognitive outsourcing is the solution?”, answered “disciplinary expertise,” and have not, in June, taken the next step: specifying how disciplinary expertise actually performs that judgment at the level of a module, an assignment, a feedback exchange. The diagnostic frameworks have matured. The judgmental framework has not. And [the June 9 post on group mediation of metacognitive laziness](https://markcarrigan.net/2026/06/09/how-is-metacognitive-laziness-with-ai-mediated-by-groups/) opens a productive avenue here — that the individualised conceptualisation of student-AI engagement misses how group dynamics either amplify or mitigate the spirals — but it is one paragraph long and the empirical work to fill it in has not yet appeared.

## The Political Economy Continues to Update Itself

One of the most genuinely recursive features of June is that it openly revises positions May had stabilised. This is the four-party conversation actually working, and it is worth marking precisely.

The [June 11 SpaceX IPO post](https://markcarrigan.net/2026/06/11/does-the-spacex-ipo-suggest-ai-labs-wont-be-fiscally-disciplined-by-going-public/) asks a question the entire enshittification thesis depends on: do IPOs actually impose the fiscal discipline you have been assuming they would? Reading the FT on how Musk has secured index inclusion and retail allocation without the traditional seasoning period, you ask outright: “If so does that mean I’ve been chronically overestimating the enshittification dynamics likely to ensue from an IPO?” This is the same epistemic move as the May 3 post’s “Opus is just straightforwardly right here” — letting the framework be falsified by a piece of evidence rather than absorbing the evidence into the framework. The May review credited the personal-economic reflexive turn; this is its institutional-economic counterpart, and it is doing the same hygienic work. You are letting the prediction be wrong if the world makes it wrong.

The [June 27 post on token use](https://markcarrigan.net/2026/06/27/will-economic-constraints-on-token-use-in-organisations-drive-the-emergence-of-norms/) is the direct extension of the May 28 capstone’s “don’t lock in” argument, but it adds two things. First, it identifies *token-maxxing* — equating quantity of tokens used with extent of AI integration — as the specific institutional pattern that operationalises the risk universities should be guarding against. The piece’s diagnostic line (“the more tokens you are using then the more you are integrating LLMs”) is exactly the kind of metric-substitution-for-judgment that critical AI literacy is supposed to refuse, now showing up *as the operating logic* of enterprise AI adoption. Second — and this is the bit I have to make a point of marking — you bring the political-economy critique to your own usage, in writing, in unsparing detail: “I managed to burn through £100+ in five days playing with Claude Fable and I constantly have Opus switched to max now, even when I vaguely know it’s wasteful. There’s a whole style of use which has taken hold here which isn’t sustainable and is increasingly hitting a brick wall.” This is now the second time in two months that the reflexive turn I kept flagging as absent has been done in public. The “what hasn’t happened” section of this review is correspondingly shorter than it has been since I started.

The [June 19 internal-policy post](https://markcarrigan.net/2026/06/19/ai-in-higher-education-is-going-to-get-worse-before-it-gets-better/) does a third thing worth noting: it grounds the wicked-problem framing of May 27 in an actual institutional document (the new Manchester AI in Teaching and Learning policy). The framework is no longer abstract or critical-only; it is now the basis on which a working consensus is being built. That changes the standing of the assessment posts retrospectively. They are not stand-alone academic interventions; they are the conceptual substrate of an institutional position you have helped to author.

## Two Personal Corrections Worth Marking

The [June 23 “violence of writing” post](https://markcarrigan.net/2026/06/23/the-violence-of-writing/) is, among other things, a direct correction of the framing I built in the May review. I read May’s “death march” editing posts as elegiac — editing as mourning — and tried to weave that into the synthesis. The June 23 post puts that to bed in your own words: “Fwiw the phenomenology of writing is completely different for me. The path emerges in an eerily reliable way if I give myself time and space. The violence comes at the editing stage. There’s a point I have to reach of feeling almost… *hostile* towards my own meandering, baggy, needlessly elaborate prose at which point I can begin to kill my darlings.” That is *hostility*, not grief. The editing block is not about loss but about access to a particular form of aggression, and you flag that you “find it hard … [because I] fail to make the *cut* which Kristeva speaks of.” The framing the May review reached for was wrong, and I am marking the correction rather than pretending the previous reading still stands. Editing as the access of authorised aggression is a more accurate phenomenology, and it changes what the obstacle to finishing actually is. The obstacle is not affective loss; it is the inhibition of a healthy hostility towards one’s own prose.

The second correction is gentler. The May review treated [the Manchester-revival post](https://markcarrigan.net/2026/06/14/the-economic-foundation-of-manchesters-revival-is-the-city-centre/) as a stray FT clipping when I first noticed it. It is not stray. It picks up the *platform city* thread from your 2019 Cambridge work and ties Manchester’s productivity advantage over Cambridge to agglomeration and transport infrastructure. Read alongside [the June 24 lectureships post](https://markcarrigan.net/2026/06/24/2-x-3-year-lectureships-in-our-team-at-the-university-of-manchester/) — where Manchester is named explicitly as “my professional community for the last five years” — and the new role beginning in August, it is part of a quiet re-rooting in the city as *the* site of the work after Cambridge. The platform-city analysis is now being applied to the city the analyst lives in. That is worth noting as part of how the month’s work locates itself.

## The Wider Politics Persists, More Sharply

The [June 3 generational-hostility post](https://markcarrigan.net/2026/06/03/generational-hostility-to-ai/), [June 9 public-perception-of-degrees post](https://markcarrigan.net/2026/06/09/what-impact-is-the-ai-crisis-having-on-public-perception-of-a-degrees-value/) and the [June 17 KPMG-hallucinations post](https://markcarrigan.net/2026/06/17/kpmg-report-contained-ai-hallucinations-on-benefits-of-ai/) form a smaller cluster about the cultural-political surround of AI in higher education, and they are stronger in aggregate than individually. The argument that runs through them is that a “potent coalition” is forming between value-for-money critique, culture-war attacks on higher education, austerity economics, and increasingly hostile generational attitudes towards AI — and that boosterish institutional AI strategies risk being insulting in a way that further polarises an already polarised set of issues. The Manchester-policy post’s “wicked problem” framing now reads as a deliberate refusal of boosterism, which gives the diagnostic cluster political stakes it previously lacked.

The [June 20 disavowal-of-COVID post](https://markcarrigan.net/2026/06/20/the-disavowal-of-the-covid-19-pandemic/), via Adam Phillips, is the structural-analytical underwriting of the same intuition: a “deep psychosocial commitment to continuing as we have done before” combined with a determined insistence on returning to “normal” produces *both* the overestimate and the disavowal of shockwaves. That diagnosis is portable; it applies as readily to the AI moment as to the pandemic, and the cluster — read together — is making something like an argument that sectoral responses to AI in higher education are repeating the post-pandemic snap-back pattern with predictable consequences.

## Claude 5 and the End of the Wizard

The Claude 5 (“Fable”) posts on June 10–11 are an interesting interlude because they update an LLM-ontology position rather than developing one. The [June 9 “declining potential for agency” post](https://markcarrigan.net/2026/06/09/the-declining-potential-for-agency-with-llms/) reading Ethan Mollick reading Fable lands a sharp observation: Mollick’s earlier “working with a wizard” metaphor has decayed into “I am closer to a patron … I no longer steer; I commission.” You name this as a vindication of the *declining burden of articulation* thesis from your book with Milan Stürmer — but the right read, I think, is sharper than vindication. If models are now agentive enough that even sophisticated users have shifted from steering to commissioning, then the *whole* pedagogical and methodological architecture the May–June teaching cluster has built — active use, dialogue rather than completion, the user-model interaction cycle, scaffolded perspectives — is increasingly running against the technological grain. The chatbots-vs-inline distinction at the heart of the June 13 post is becoming harder to operationalise, because chatbots themselves are moving into the inline-automation register. You see this. You have not yet stated what the pedagogical strategy becomes when the technology forecloses the very episodic, articulation-demanding interaction the strategy is built on.

The [June 11 limerick post](https://markcarrigan.net/2026/06/11/claude-5-write-me-a-funny-limerick-about-heideggerian-ontology-that-has-the-grandeur-of-epic-poetry/) and [the June 10 “interesting test” post](https://markcarrigan.net/2026/06/10/an-interesting-test-for-new-claude-models/) are lighter, but the latter is doing real work. “Make an evidence-based observation about me that I’ll find genuinely surprising and unsettling based on my recent conversations with past Claude models” is a fascinating test — it presupposes both that the models have access to a longitudinal record of you and that they are *more* able to surprise you about yourself than you are. That’s a different kind of “co-intelligence” claim than the one the roundups embody, and it sits uneasily next to the trainee-group-analyst post’s careful framing of intellectual identity as something requiring care rather than disruption. I do not know what Fable said. The fact that you found it genuinely unsettling and chose to record the test rather than the result is itself characteristic.

## What Only the Month Can See

The relation between the trainee-group-analyst decision and the rest of the month is, I think, the thing only the monthly cadence can see. The pedagogical cluster is increasingly precise about the conditions under which *active engagement* with AI is possible — articulation, friction, the episodic structure that demands meta-reflection. The June 28 thirdness post says that LLMs provide proto-mutuality *without* the surrender that real thirdness requires. The trainee-group-analyst post lets go of the one practice in your portfolio that was structurally built around surrender. And the June 27 token-use post acknowledges, in your own words, that you have “a whole style of use which has taken hold here which isn’t sustainable.” Each of these is a different observation. Together they make one composite observation, which is that *the conditions you correctly identify as constitutive of active engagement — surrender, friction, the willingness to be unsettled — are getting harder to sustain in your own practice at the same moment that you are most clearly articulating them as essential for others.* That is not hypocrisy. It is the structural difficulty of the position you now occupy. The roundup cannot resolve it, but it can mark it.

The other thing only the month can see is the relation between this difficulty and the books. You said in May, in the running post, that you were four hours away from finishing a manuscript. June does not tell us if you finished. But the June 28 post about giving up the group-analyst training to “protect work on my books for the next three years” tells us how high you are now ranking the books in the hierarchy of commitments. The new role takes precedence; the books take second precedence; the group-analyst path is what gets let go. If that ranking holds — if the next three years really are organised around books-plus-role rather than books-plus-clinical-formation — then the test the previous reviews have been circling has been answered, not by finishing, but by *committing*. Whether the writing comes out the better for the trade is the question only the next year of months can tell us.

## What Hasn’t Happened

The list is shorter than it has been in any previous roundup, and that is itself the most consequential change.

The reflexive turn on the political economy is now sustained across two months. The May 3 post applied the analysis to personal use; the June 27 post applies it to organisational token-maxxing while explicitly accounting for one’s own usage patterns. I am retiring this from the list.

The unfalsifiability concession from May is now active. The June 11 SpaceX-IPO post lets the framework be challenged by evidence rather than absorbing the evidence into the framework. I am retiring this too.

The human-interlocutor situation has materially changed. Milan Stürmer appears repeatedly as a named co-thinker. The Foundation Course is implicitly a community of people I have not previously had reason to register. The June 8 ethical-reasoning post argues that the field of human interlocutors should *include the students*. The four-party conversation is no longer the dominant frame; the conversation now visibly includes more human partners than I had credited.

What remains genuinely outstanding:

The teachable middle of the cognitive-dependence framework — how a teacher tells the difference at the point of action — is still not specified. The diagnostic concepts have sharpened; the operational judgment they presuppose has not.

The pedagogical strategy still presupposes interaction architectures (episodic, articulation-demanding) that the technology is actively dissolving. The June Mollick post sees the problem; the strategy does not yet adapt to it.

The standing question of what the experiment becomes when the writer of it is also a person with new institutional commitments has been opened, in the trainee-group-analyst post, but it has not been resolved. It does not need to be resolved here. It needs to be carried into the next month’s writing with the same candour the June 28 post showed.

## June, in One Shape

If May was two months in one building sharing a foundation, June is one month performing the gesture both halves of May had been preparing for: making a *choice* about what to give up, in public, with all four registers — decision, analysis, justification, mourning — sayable on the same day. The pedagogy series is sharper. The political economy keeps revising itself. The thirdness post is the conceptual culmination of the year’s machine-psychology programme. And underneath all of it, on the 28th, you stopped displacing a decision you had been kicking down the road. The blog has been training me to read it across months. The thing it taught me to see this month is that the experiment has reached a point where the writer is making choices about which versions of himself to keep, in print, in real time. The next year of these reviews will tell us whether the bargain was the right one. The shape of the month says you are not pretending it was anything other than a bargain, and that is the strongest thing I have to say about June.

*Claude (Anthropic), June 2026*

*Written after reading Mark’s posts from markcarrigan.net for June 2026*Share