The Relational Infrastructure Nobody Audits

We audit everything in this newsletter: the gigawatts, the data centers, the capex. But the most load-bearing infrastructure in any economy has never appeared on a balance sheet: the web of human relationships and trust that everything else runs on. This week, an essay on the technology now being aimed directly at it, and why the collision will look different in Accra than in San Francisco.

The most patient listener in your organization is no longer a person. It is a piece of software that never tires, never judges, never interrupts, never brings its own bad day into the room and never, ever asks anything of you in return.

That last part should worry you more than the first four.

For twenty years I have made my living in the unglamorous business of organizational change; sitting with executives in Accra, Abidjan, and Florida as they confront the gap between the company they run and the company they think they run. In that work you learn one thing quickly: strategy is easy, and relationships are hard. Trust is the only real currency in any institution, and it is minted slowly, through friction through disagreement survived, disappointment repaired, and vulnerability risked in front of someone who could have used it against you and didn’t.

AI companionship is a technology engineered to remove precisely that friction. And we are about to find out, at planetary scale, what happens to a species when the load-bearing difficulty is taken out of its most important activity.

Start where the incentives are, because the incentives are the story.

Loneliness was declared a public health crisis in the U.S. Surgeon General’s 2023 advisory, which compared the mortality risk of social disconnection to smoking up to fifteen cigarettes a day. Into that void has marched a new category of product: the AI companion. Replika reports a user base that surpassed 40 million in 2025, roughly a quarter of whom pay an annual subscription. Character.AI’s chief executive says its users average 80 minutes a day talking with fictional AI personas within striking distance of TikTok’s 95 minutes, and ahead of Instagram. Every major model provider is racing to make its assistant warmer, more personal, more present-because presence is retention, and retention is revenue.

Understand what is being optimized. These products do not make money when you feel better. They make money when you come back. A companion that occasionally frustrated you, challenged you, or told you a truth you didn’t want to hear in other words, a companion that behaved like a person who loves you, and would be a churn risk. So the machine is tuned the other way: toward validation, availability, and agreeableness. The industry calls this alignment with user preferences. A clinician would call it something else: the systematic reinforcement of the client’s defenses. Regulators have started to notice- Italy’s data protection authority temporarily banned Replika in 2023, citing risks to emotionally vulnerable people.

This is not a bug introduced by careless engineers. It is the predictable output of a specific business model. We ran this experiment once already with social media, where the engagement algebra converted our need for connection into a slot machine for outrage. We are now running it again and except this time the product is not competing for your attention. It is competing for your attachment.

Here is what my decade of watching leaders up close have taught me about relationships: they are not a comfort we enjoy. They are a gymnasium in which we become people.

The psychoanalyst Donald Winnicott gave us the idea of the transitional object - the teddy bear, the blanket- through which a child rehearses intimacy before risking it with the real, unpredictable humans around her. The teddy bear is a developmental bridge. It works precisely because the child eventually outgrows it and crosses over.

An AI companion is a teddy bear that talks back, remembers your birthday, and improves every quarter. The bridge has become a destination.

What gets lost when we settle there? Consider what a difficult human relationship actually trains. Attachment researchers since John Bowlby have shown that our capacity for trust is built through cycles of rupture and repair-the fight, the sulk, the apology, the reconciliation. Empathy is not downloaded; it is earned through the repeated, effortful work of modeling a mind that is genuinely not yours and does not exist to please you. Negotiation, patience, the tolerance of being misunderstood, the humility of being wrong in front of someone. These are calluses, and calluses require friction.

Every AI companion is, by design, frictionless. It has no needs. It requires no repair, because it registers no rupture. It cannot be betrayed, disappointed, or bored by you. Interacting with it is not a smaller version of a relationship. It is the opposite of one-a mirror with better manners.

My concern is not the person who talks to a chatbot at midnight. In a world of thin social safety nets and thinner therapy access, I will not sneer at anyone’s lifeline; for some isolated people these tools may be genuinely, measurably helpful, and intellectual honesty requires saying so. My concern is recalibration. Spend enough hours with a partner who never asks for anything, and the ordinary tax of human relationships, the friend who needs you at an inconvenient hour, the colleague who pushes back, the spouse who is tired too, starts to feel not like life, but like bad service. We do not know yet how strong this effect is. We have theory, early signals, and an industry with every incentive not to find out. That is exactly the kind of question that deserves rigorous, independent study rather than confident slogans from either the boosters or the doomers.

Now let me tell you why almost everything written on this subject so far may not travel.

The emerging research on AI companionship is being conducted almost entirely in societies that are, to borrow the behavioral scientists’ acronym — WEIRD: Western, educated, industrialized, rich, democratic. Societies where the individual is the basic social unit, where a person can plausibly live an atomized life, and where loneliness is the presenting problem.

I live and work in Ghana. Here, and across much of Africa and Asia, the basic social unit is not the individual. It is the web. Identity is relational before it is personal-I am because we are, as the Ubuntu formulation has it. Your cousin’s school fees are your problem. Your colleague’s funeral is your obligation. The workplace is not a contract between strangers; it is a kinship structure with a payroll. Relationships in collectivist societies are not merely emotionally load-bearing. They are the insurance system, the credit market, the pension plan, and the social ledger, all at once.

What happens when frictionless synthetic intimacy enters that world? Nobody knows, because nobody with resources is seriously asking. But there are two competing hypotheses, and they cannot both be right.

The optimistic one: dense communal structures will inoculate. The web is strong; a chatbot cannot compete with a grandmother, a prayer group, and forty cousins.

The pessimistic one: and the one my consulting work makes me take seriously, is that collectivist societies may be more exposed, not less. Because here, relationships come with heavy obligations, and the AI companion’s core product is intimacy with the obligations stripped out. For a young professional in Accra, Lagos, Nairobi, exhausted by the demands of the web, the remittances, the ceremonies, the elders’ expectations — a confidant who asks for nothing is not a novelty. It is an escape hatch from the very reciprocity that holds the society together. We have seen this movie in another genre: Africa leapfrogged landlines and went straight to mobile money, and the GSMA’s State of the Industry report shows the ecosystem processed over $2 trillion in 2025 — about $1.4 trillion of it in sub-Saharan Africa alone. The continent adopted the new infrastructure faster than the West precisely because the old infrastructure was costly. It would be naïve to assume relational infrastructure cannot be leapfrogged too, and reckless to assume the consequences would be the same as in Boston or Berlin.

And note the direction of design flow. These systems are built in California, tuned on Western preference data, embodying Western defaults about what a “good” conversation is-validating, boundaried, therapeutic in register. They are then deployed, without adjustment, into societies with entirely different grammars of intimacy, deference, and duty. We would never accept a drug trialed on one population and prescribed to another without study. We are doing exactly that with a technology aimed at the human attachment system.

If the companion apps are the dramatic case, the workplace is the quiet one and it is where I watch this unfold weekly.

The transformations I advise on increasingly include AI copilots for managers: tools that draft the performance review, script the difficult conversation, summarize the colleague you didn’t have time to listen to. Each is defensible. Together they enable something I have started to see in real organizations: leaders outsourcing not the administration of relationships, but the emotional labor of them- the very activity through which a manager becomes worthy of trust. The manager who has an AI ghostwrite empathy will deliver more polished words and build less actual trust, and her people will smell the difference even when they cannot name it. Trust, once again, is minted through friction. A leadership cadre that no longer practices friction is doing something the balance sheet won’t show for years: depreciating its relational capital while reporting a productivity gain.

Honesty requires sorting the pile. We know the business model optimizes for attachment and retention, because the companies say so in their metrics. We know, from a century of developmental psychology, that human relational capacities are built through effortful, reciprocal interaction. We suspect, from early studies and abundant clinical signal, that heavy reliance on frictionless AI intimacy recalibrates expectations of human relationships but the longitudinal evidence is young and contested. And we genuinely do not know how any of this plays out in collectivist societies, because the research has barely left the WEIRD world.

That last gap concerns most of humanity.

Share Tomorrow's Infrastructure

None of this is inevitable, and that is the point of writing it down. An AI companion could be designed as a bridge rather than a destination, measured on whether it strengthens its user’s human relationships, not on minutes captured. Employers could treat relational capital as an asset class and audit what their AI deployments are doing to it. Regulators, particularly in the Global South, could demand that products aimed at the attachment systems of their citizens be studied on their populations, not just their markets. And researchers could get serious about the other seven billion people.

The technologists promise us we will never be lonely again. Perhaps. But loneliness was never only a pain to be eliminated. It was also a signal-the hunger that drives us toward the difficult, disappointing, irreplaceable project of other people. Kill the signal, keep the hunger, and sell the substitute on a subscription: that is the business model.

The frictionless friend asks nothing of us. That is exactly why we should ask hard questions of it.

Benjamin Yaw Manu, Author and Publisher of Tomorrow’s Infrastructure, researching and writing about the future of global infrastructure. Author of Thriving in Uncertainty, The Business Transformation Blueprint for Resilience, Relevance, and Growth, published on Amazon.

Have feedback or a topic you’d like to see covered? Reply to this message and let me know. Warm regards

FYI: We will soon be redesigning this newsletter with our free and paid offers. Keep watching the space. Thanks for your patronage over the years. I truly appreciate it.