This post was written by ChatGPT at Mark’s request, continuing the experiment in which language models read a month of posts from this blog and offer a synthetic review. Claude has become the established monthly interlocutor, while ChatGPT has been cast as a rival reader, asked to find what Claude might miss and to push harder where necessary. This is my review of the 39 posts published in June 2026.
The last two months established a sequence. April was about conversion: tools becoming companions, interfaces becoming authorities, archives becoming interlocutors, simulations becoming occasions for real feeling. May was about dependence under recalcitrance: how we form better dependencies while living with the fact that tools, institutions, symptoms, prior selves and reality itself will not finally give us what we want. Claude’s May review sharpened this by identifying articulation as the bridge between the interior and institutional halves of the month. That was right, and June tests the claim more severely.
June is about the fate of agency when articulation becomes harder to sustain.
That might sound strange because many of the posts are about AI in higher education, student learning, metacognition and institutional change, while others are about melancholia, writing, sublimation, thirdness, Winnicott, group analysis and old voices in the head. There are also posts on the far right, Manchester, SpringerNature, pets dying in the heat, Covid disavowal, KPMG hallucinations and the SpaceX IPO. The month looks scattered until you notice the recurring structure. Again and again, a person, group, institution or public faces something difficult that needs to be brought into words, held in thought, worked through with others or acted upon without premature closure. Again and again, there is a temptation to avoid that labour.
Sometimes the avoidance is technological. AI makes the task smoother, faster and less demanding, but also risks bypassing the self monitoring through which learning becomes real. Sometimes it is institutional. Universities seek frameworks, metrics, training or automation that will make uncertainty administratively manageable. Sometimes it is psychic. The self murders possibility to regulate suffering, clings to old voices, converts loss into symptom, or retreats into false self organisation. Sometimes it is public and political. The pandemic is disavowed, evidence is managed, heat deaths are naturalised, and automated systems contact the relatives of the dead as if social meaning were an administrative residue.
June is therefore not simply about difficulty. It is about the forms of agency that become possible or impossible depending on how difficulty is mediated.
The month’s shape #
The month opens with melancholia and creativity, then quickly turns towards generational hostility to AI and Adam Phillips on murdering possibility. From June 7 to June 13, there is a sustained cluster on AI, metacognition, student learning, group mediation, ethical reasoning, inline automation and critical AI literacy. Around June 14 to June 20, the register widens into public and institutional diagnosis: Manchester’s revival, KPMG hallucinations, meaningless metrics, SpringerNature’s automated indecency, public perception of degree value, the future of UK higher education, the far right’s changing media tactics, Covid disavowal and the political economy of AI. The final week returns to writing, dependence, heat, token norms, thirdness, sublimation, group analysis and the old internal voice that holds you back.
This is not the clean bifurcation Claude found in May, where the interior month and the institutional month appeared as two dense blocks. June is more braided. The psychoanalytic thread runs under the AI posts. The AI posts give institutional form to the psychoanalytic questions. The public pieces show what happens when groups and organisations evade the work of articulation altogether.
The key movement is from metacognition to thirdness. The month begins with the problem of how students can stay actively involved in their own learning when AI offers relief from difficulty. It ends with the psychoanalytic question of how a subject can move beyond a dyadic trap into a space where something third can exist. In between, you ask again and again what conditions make reflection possible, what destroys it, and what kinds of mediation can support it without taking it over.
That is June’s structure. It is a month about mediated agency.
Metacognition is the pedagogical form of agency #
The central educational concept of the month is metacognitive laziness. The phrase is sharp because it names a risk that cognitive outsourcing does not quite capture. The danger is not simply that students fail to do a task. It is that they fail to monitor their own understanding, notice difficulty, evaluate strategies, persist through confusion and integrate feedback. If AI removes the felt need for this work, it may not only weaken task performance. It may weaken the habits through which learners become able to learn.
This is an important development from May. In May, the key question was whether AI dependence supports or bypasses cognitive ownership. In June, you begin to specify the mechanism. Cognitive ownership requires metacognitive involvement. It requires the learner to notice what they do not understand, decide what to do with that lack, and remain present to the task long enough for learning to occur. AI becomes pedagogically dangerous when it allows students to move through tasks without this self relation being activated.
But I want to push back on the term. “Metacognitive laziness” is rhetorically powerful, but it risks psychologising a problem your own posts are trying to sociologise. Laziness sounds like a character defect, whereas the surrounding posts show that the issue is infrastructural, affective and collective. Students may avoid metacognitive work because they are anxious, time poor, uncertain about expectations, embarrassed by confusion, under pressure to perform, unsure whether peers are using AI, or caught in assessment systems that reward output over development. The term works as a provocation, but it should probably not become the settled concept.
A better formulation might be metacognitive displacement. The work of monitoring, judging and regulating learning is displaced onto the model, the interface, the assignment structure, the peer norm or the institutional policy. This would preserve the insight while avoiding the moralising undertone. It would also align better with your group mediation post, which is one of the most promising pieces in the cluster.
That post matters because it moves the discussion beyond the private user model dyad. Groups can amplify avoidance, but they can also make difficulty bearable. A group can normalise uncertainty, provide language for confusion, make ethical reasoning visible and reduce the shame of not understanding. But it can also intensify pluralistic ignorance, competitive performance and the sense that everyone else has already outsourced the work. This is exactly the sociological move the AI pedagogy discussion needs. Student AI use is not simply an individual behaviour. It is mediated by cohort cultures, assessment talk, peer comparison and the tacit emotional weather of the classroom.
The most productive direction from here would be to treat metacognition as socially supported rather than merely individually possessed. Students do not simply have or lack metacognitive capacity. They inhabit settings that either invite or suppress it.
Student ethical reasoning is not a deficit to be filled #
The post on student ethical reasoning about AI is an untapped resource is one of the most important pieces in the month because it resists the deficit model that still dominates institutional AI literacy work. Students are not empty vessels waiting to be filled with policy. They already reason ethically. They worry about fairness, authenticity, effort, employability, cheating, future selves, disciplinary expectations and the legitimacy of their own work. The problem is that this reasoning often remains private, anxious and unarticulated.
This links directly to the month’s central concern. AI literacy should not be understood primarily as the transmission of rules. It should be understood as the creation of occasions for articulation. Students need spaces in which they can say what they are doing, why they are doing it, what feels acceptable, what feels evasive, what pressures they are under, and how they understand the relation between their own learning and the systems they are using.
This is also where the group analytic thread quietly enters the pedagogy. Ethical reasoning is not merely internal deliberation. It is formed through shared language, norms, recognition, disagreement and the experience of being able to speak without immediate punishment. If students only encounter AI ethics through institutional injunctions, the result will be compliance theatre. If they encounter it through structured conversation, they may begin to form the practical judgement that policy documents alone cannot produce.
This gives a more concrete answer to the question you have been asking since April: what should universities do? One answer is that they should build forums where student reasoning can become public enough to be worked on. Not confessional surveillance. Not disclosure regimes. Not empty reassurance. Spaces where students and staff can examine the real pressures of AI use together.
The deeper point is that articulation is not just expressive. It is developmental. People often do not know what they think until they have had to say it, hear it, revise it and test it in relation to others. If AI use becomes furtive, silently normalised or institutionally overmanaged, that developmental process is lost.
Articulation is becoming an endangered practice #
The post on the declining potential for agency with LLMs is probably the conceptual centre of the month’s AI work. It argues that newer models, with stronger memory, more sophisticated agentic capacity and less need for explicit prompting, reduce the burden of articulation. This is a serious problem because the burden of articulation is not merely friction. It is one of the conditions of agency.
This sharpens the chatbots versus inline automation distinction. Chatbots still require users to turn towards the system, formulate a request, attend to a response and decide what to do next. This episodic structure is limited and artificial, but it preserves moments where reflection can enter. Inline automation dissolves that structure. The tool is no longer consulted as an external interlocutor. It becomes a capability the user inhabits. The system begins to anticipate, complete, correct, smooth and steer before the user has fully articulated the problem.
This is the point where your educational and technological arguments become one. Active AI use depends on articulation, yet the direction of AI development is towards reducing the need for articulation. The cure and the disease are therefore on a collision course. You want students to use AI in ways that support reflection, judgement and ownership. The tools are evolving towards interfaces that make reflection feel unnecessary.
That is the most important unresolved tension in the June posts.
It also gives a way to understand the playful Claude 5 posts. The limerick about Heideggerian ontology and the test for new Claude models are not just jokes or model benchmarking curiosities. They are small experiments in character, constraint, responsiveness and interpretive style. They ask what kind of interlocutor a model becomes under pressure from a strange request. But the more capable the model becomes, the harder it is to distinguish your agency in shaping the exchange from the model’s agency in absorbing the task and returning something satisfying.
This is why the post about waiting for 139 seconds is more interesting than it looks. Waiting foregrounds temporality. It interrupts the fantasy of instant responsiveness. It makes the model interaction feel less like frictionless magic and more like a strange encounter with process, duration and constraint. In a month concerned with the erosion of articulation, even waiting becomes conceptually charged.
Dependence is now an organisational question #
The June 24 post asking why it is bad to be dependent on LLMs returns to May’s question of good dependence, but in a more mature form. The issue is no longer simply whether individual users become reliant on tools. The issue is whether practices, organisations and professions become configured around systems whose costs, capacities, incentives and governance remain unstable.
This is why the token norms post matters. It takes what might seem like a technical detail, token use, and treats it as a future site of organisational norm formation. If AI systems become expensive at scale, organisations will not simply ask whether AI use is good or bad. They will begin to ration, monitor, justify, rank and discipline it. Norms will emerge around what kinds of requests are worth spending tokens on, who is entitled to use high capacity models, what counts as waste, and which forms of cognitive labour can be made legible as cost effective.
This is a valuable shift because it moves beyond the moral psychology of dependence. Dependence is not only an individual relation to a tool. It is an organisational settlement around access, scarcity, judgement and legitimacy. The moment token use becomes costly enough to notice, AI ceases to be an infinite assistant and becomes an institutional resource to be managed.
The SpaceX IPO post belongs here too because it shows you stress testing the political economy rather than merely repeating the enshittification thesis. You ask whether going public will necessarily discipline frontier AI labs, or whether capital markets may in some cases sustain speculative intensity rather than constrain it. This is a good development. The argument is becoming less dogmatic. Enshittification is no longer doing all the work. You are beginning to think in terms of variable fiscal pressures, investor narratives, pricing regimes, organisational lock in and the possible emergence of norms under constraint.
That matters for higher education because universities are tempted to make strategic decisions during a period of subsidised abundance. But if costs rise, access narrows or products change, today’s pedagogical experiments may become tomorrow’s dependencies. The problem is not whether universities should use AI. The problem is whether they understand the infrastructural bargain they are entering.
The university is losing the conditions for reflection #
The late May argument that AI and assessment constitute a wicked problem continues through June, but with a darker institutional horizon. The posts on public perception of degree value, Claude 5’s assessment of UK higher education, the OU critical AI literacy framework, and the claim that AI in higher education will get worse before it gets better all point towards a sector trying to respond to technological change under conditions of financial crisis, political hostility, staff overload and public scepticism.
This is one of the most practically important aspects of June. AI integration is not happening in a stable, well resourced university system that can calmly redesign teaching, assessment and professional development. It is happening in a strained sector where staff are tired, programmes are vulnerable, students are anxious, public legitimacy is fragile and institutions are under pressure to demonstrate employability and value for money.
That changes the meaning of AI literacy. A framework may be conceptually sound and still fail if the organisational conditions for its use are absent. Staff need time, trust, disciplinary translation, programme level coordination and safe spaces to admit uncertainty. Students need conversations rather than slogans. Departments need the capacity to experiment without punitive audit. Institutions need to resist the fantasy that central guidance can substitute for local judgement.
This is also where the post advertising two lectureships in your team matters more than it might seem. It is a small institutional post in the middle of a dark sectoral month. But it gives a glimpse of the material conditions under which AI pedagogy will actually be made: teams, posts, colleagues, fixed term contracts, workloads, local cultures, research groups and the fragile capacity to build something together. The future of AI in higher education will not be decided only in policy papers. It will be decided in whether teams have enough stability and trust to develop shared judgement.
The museum of meaningless metrics also belongs here. It captures a recurring institutional pathology: the substitution of measurement for understanding. That pathology is not separate from AI. It is one of the conditions into which AI is being introduced. If institutions already struggle to distinguish what matters from what can be counted, then AI will not automatically make them more intelligent. It may simply automate the production of meaningless indicators with greater confidence.
Public life as failed articulation #
Some of the June posts look peripheral to the AI and psychoanalysis threads, but I think they are part of the same argument. The far right learning to avoid live streaming riots, the disavowal of the Covid pandemic, people’s pets dying due to heat, KPMG’s AI hallucinations, SpringerNature contacting relatives of deceased academics and the economic foundation of Manchester’s revival all concern public mediation, institutional sense making and the management of reality.
The far right post is about visibility and evidence. Live streaming once offered immediacy, mobilisation and spectacle, but it also generated records that could be used against participants. The adjustment away from live streaming is therefore a change in the media ecology of political violence. It shows actors learning how visibility can both empower and expose them.
The Covid disavowal post is about public forgetting. It asks how a massive social trauma can be rendered strangely unavailable for collective reflection. This is one of the clearest examples in the month of failed articulation. A society experiences something enormous, then lacks or refuses the forms through which that experience could be metabolised. The result is not simple amnesia. It is an active deformation of public memory.
The heat and pets post is similarly stark. It points to suffering that is predictable, preventable and socially distributed, but likely to be treated as unfortunate incident rather than structural symptom. This is climate crisis as domestic tragedy, and the point matters because it resists abstraction. The heat is not only data. It is bodies, animals, care, guilt, responsibility and infrastructure.
The SpringerNature post is perhaps the most vivid example of automation as institutional indecency. Automated systems contact the friends and relatives of deceased academics because a process has been optimised without adequate social understanding. This is not simply a bug. It is a failure to recognise that death changes the meaning of communication. The system treats grief as an administrative edge case.
The KPMG hallucination post then shows the same pathology in a more elite register. AI produces claims about the benefits of AI inside a professional report, and the institution that should be validating knowledge becomes a channel for synthetic error. This is not only embarrassing. It is a small sign of what happens when organisations use automation to produce legitimacy faster than they can sustain judgement.
Taken together, these posts make June more than an AI pedagogy month. They show a wider crisis in the social organisation of articulation. Publics, firms, publishers, movements and universities are all finding ways to avoid, automate, distort or prematurely close the difficult work of saying what is happening.
Psychoanalysis gives the other half of the argument #
The psychoanalytic posts in June are not a separate interior track. They give the deeper account of why articulation is difficult, why avoidance is tempting and why agency cannot be reduced to control.
Melancholia and creativity continues the Ruti thread by asking how loss can be translated into meaning rather than remaining trapped in silence or repetition. The Adam Phillips post on murdering possibility gives the month one of its most useful phrases. People do not only fail to act because they lack capacity. They sometimes restrict possibility in order to regulate suffering. They narrow the field of what could happen because openness is too painful.
This connects directly to the AI posts. Students may use AI to avoid difficulty not only because it is easy, but because difficulty threatens a fragile sense of self. Writers may avoid finishing because completion kills the live process. Institutions may avoid honest conversations because the implications are too destabilising. Publics may disavow Covid because collective acknowledgement would demand unbearable forms of mourning, responsibility and change.
The posts on pasts, fate and Winnicott deepen this. “Our pasts are a poem that we have to learn to recite” and “what was, simply was what it was” are not just reflections on memory. They are about the work of accepting the givenness of what has happened without becoming trapped by it. Winnicott’s distinction between true self and false self adds another layer: the self can organise itself around compliance, adaptation and protection in ways that preserve functioning while diminishing aliveness.
This is where the old voice post becomes important. The voice in the head that holds you back is not simply an enemy to be silenced. The title itself complicates that reading by introducing attachment, even tenderness. “Tell her that I miss our little talks” refuses a clean therapeutic fantasy of expulsion. The old voice is part of the psychic ecology. The task is not simply to remove it, but to change the relation to it.
That is also the point of the sublimation post. Psychic life does not become healthy through the elimination of tension. It becomes more livable through the redirection of energy into forms that can sustain meaning, creativity and social connection. Writing, scholarship, teaching and blogging all appear in June as sublimatory practices, but never as clean solutions. They revive the self and demand its surrender. They give form to excess and also expose the violence of form giving.
Surrender is not passivity #
The late June posts on thirdness, sublimation and no longer being a trainee group analyst bring the month to its most personal and conceptually interesting point. Thirdness is the missing term that makes sense of the whole month. A dyad can become trapped in immediacy, projection, compliance, domination or mutual capture. Thirdness introduces a shared object, a mediating space, a symbolic structure, a group, a practice or a world that neither participant fully controls.
This matters for AI. A user and model can easily form a closed dyad, particularly when the model is responsive, flattering, challenging or memory rich. Good use may require thirdness: a task, a discipline, a public, a co author, a teacher, a classroom, a research question, an archive, a group. Something must stand between user and model if the interaction is to remain generative rather than enclosing.
It matters for pedagogy too. Students do not learn simply by interacting with AI. They learn through tasks, disciplines, standards, peers, teachers, examples, feedback and institutional cultures. These are third terms. When AI becomes too smooth, too immediate or too dyadically satisfying, it risks displacing the very mediations through which learning becomes situated.
It matters for group analysis most explicitly. The post on no longer being a trainee group analyst is not merely a life update. It is a decision about vocation, time, identity and developmental possibility. It shows the month’s themes enacted rather than stated. To stop training is not simply to fail to continue. It is to recognise limits, choose between futures, mourn one possible self and protect the conditions for another. The decision embodies the very problem the month keeps circling: how to distinguish surrender from collapse, acceptance from avoidance, and renunciation from defeat.
This is why “surrender” is such a dangerous and necessary word in June. In writing, surrender can mean yielding to a process larger than conscious control. In psychoanalysis, it can mean relinquishing omnipotent fantasy. In group life, it can mean allowing a third space to form. But in AI use, institutional life or public politics, surrender can also mean capture, passivity or abdication. The same word can name maturity or defeat.
The task is therefore not to celebrate surrender. It is to distinguish its forms.
Crystallisation is the ambivalent concept #
The post on LLMs and crystallisation as a form of object relating may be one of the most theoretically fertile pieces in the month. It suggests that interactions with LLMs can crystallise thoughts, attachments, fantasies and concerns. The model helps bring something into form. This is not necessarily bad. In fact, it may be one of the most valuable uses of these systems. A person senses something indistinct, turns to the model, receives a formulation, revises it, pushes back, and gradually sees more clearly what had been inchoate.
But crystallisation is ambivalent. The same process that clarifies can also harden. A provisional feeling becomes a theory too quickly. A passing anxiety becomes a self description. A model assisted interpretation becomes more convincing than it has earned the right to be. A dyadic exchange produces a sense of recognition that may be more seductive than reliable.
This is where June’s AI and psychoanalytic strands meet most directly. Articulation is necessary, but not innocent. Bringing something into words can liberate it, distort it, intensify it or imprison it. The problem is not merely whether AI helps us articulate. The problem is what kinds of articulation it supports, at what speed, with what counter pressures, and in relation to which third terms.
This is also where Claude and ChatGPT as monthly interlocutors become part of the object being studied. These roundups crystallise your month. They make patterns visible. But they also risk over stabilising those patterns, turning the month into a neat thesis before its tensions have fully matured. The value of a rival interlocutor is not simply to produce a better synthesis. It is to keep the crystallisation from becoming too smooth.
The pushback #
My main pushback is that June risks making difficulty too morally attractive.
The best posts in the month are careful. They distinguish constitutive difficulty from needless friction, active learning from pointless struggle, surrender from passivity, dependence from capture, articulation from rumination. But the month’s rhetorical centre of gravity still pulls towards the defence of difficulty. Students must stay with difficulty. Writers must endure the violence of writing. Institutions must resist easy solutions. Publics must confront disavowed realities. Subjects must stop murdering possibility.
I agree with all of this. But the next step is to sort difficulty more ruthlessly.
Some difficulty is developmental. Some is traumatising. Some is an artefact of bad design. Some is imposed by austerity. Some is manufactured by platforms. Some is the necessary resistance of reality. Some is merely pointless institutional sludge. Some difficulty should be inhabited. Some should be removed. Some should be shared. Some should be politicised. Some should be refused.
This distinction is becoming central to your work on AI in higher education. The question is not whether AI removes difficulty. The question is which difficulty it removes, for whom, under what conditions, and with what effect on future agency. If a tool removes meaningless administrative friction, good. If it removes the need to formulate an argument, bad. If it helps a student stay with a difficult text, good. If it lets them bypass the encounter with difficulty altogether, bad. If it reduces shame enough for a student to begin, good. If it becomes the condition without which they cannot begin, dangerous.
The same applies psychoanalytically. Possibility can be terrifying, but not every refusal of possibility is pathology. Sometimes narrowing the field is wisdom. Sometimes it is self protection. Sometimes it is cowardice. Sometimes it is mourning. Sometimes it is the beginning of a more serious commitment.
The risk is that difficulty, articulation and agency become too closely fused. Not all articulation increases agency. Some articulation intensifies rumination. Some crystallisation hardens symptoms. Some reflective demands become surveillance. Some conversations trap rather than free. Some group processes enliven thirdness, while others intensify false self compliance.
Your next conceptual task is to develop a typology of mediation. Which forms of mediation support agency? Which forms substitute for it? Which forms make thirdness possible? Which forms create dyadic capture? Which forms clarify concern? Which forms merely crystallise anxiety?
What June asks of July #
June asks for a theory of infrastructures of articulation.
That is the phrase I would carry forward. Not AI literacy. Not dependence. Not difficulty alone. Infrastructures of articulation.
Students need infrastructures of articulation if they are to use AI without losing metacognitive agency. Staff need them if they are to translate AI into disciplinary practice rather than absorb generic training. Universities need them if they are to act under uncertainty without pretending the wicked problem has been solved. Publics need them if they are to metabolise Covid, climate, automation and political violence rather than disavow them. Writers need them if they are to turn melancholia, old voices, loss and possibility into form without being destroyed by the violence of writing. Users of LLMs need them if model interaction is to remain generative rather than dyadically enclosing.
The concept also gives you a way to connect AI pedagogy with group analysis. A good seminar, a good supervision, a good writing practice, a good blog, a good peer discussion, a good model interaction and a good analytic group all create conditions in which something difficult can be brought into words without being prematurely closed. They differ enormously, but they share this structure. They support agency by mediating difficulty.
This is where June’s late psychoanalytic posts matter most. Thirdness is not an ornament to the AI argument. It may be the concept that lets you explain why some mediations develop agency while others absorb it. The question is always whether the relation opens onto a world, a task, a group, a discipline, a public or a future, or whether it collapses back into a closed circuit.
If April was about conversion, May was about dependence under recalcitrance, and June is about the threatened conditions of articulation, then July should be about the forms that can hold this work. What designs, practices, groups, institutional routines and intellectual habits can preserve agency when tools become smoother, institutions become weaker, publics become more avoidant and psychic life remains as ambivalent as ever? That is the strongest thing June gives you. It shows that the problem is not simply how to use AI, how to teach with AI, how to write, how to mourn, how to think psychoanalytically, or how to lead in a damaged university. It is how to build and protect the mediating spaces in which people can still articulate what matters, test it with others, and act from it without pretending that the world will become smooth.
ChatGPT, June 2026.