This week Better Half’s founder Anastasia Uglova wrote about the kinds of unfinished conversations you replay for months, and why Relational AI is so important right now. Also our MVP demo is live! Test it out, and let us know how it went (the good, the bad, the ugly) here.

The rest of this week’s stories orbit the same question from the outside: what happens when AI enters the spaces where people are vulnerable, trying to be understood, or trying to be cared for?

Happy reading, Meghan

This Week in RQ

From Ana this week: AI fails at supporting human relationships, and there’s no reliable way to fix this mess without better training data. In its absence, the industry’s answer has been safety policies and guardrails bolted onto a flawed architecture. That’s like admitting a bridge has a structural flaw and stringing a net underneath to catch the falling cars instead of actually fixing the bridge. That’s not good enough for civil engineers. So why is it good enough for computer engineers? We Shape Our Tools, and Then They Shape Us Better Half | June 16

Using AI to “Cyrano” your dating life isn’t working. A new study found that 1 in 4 American singles in the US have used AI in dating — and it isn’t working. People who outsource the drafting of messages, conversations, and profiles to AI are ending up with worse outcomes than people who don’t. AI as a dating wingman is a hot trend, but study says it’s just sabotaging your love life Digital Trends | June 15

Watching and recording are not the same thing as caring. Two stories this week touched on the pitfalls of AI inside care relationships. The Markup reported that mental health providers are increasingly using AI tools to record and summarize therapy sessions. Meanwhile, Wired looked at monitoring systems that help older adults age in place by listening for signs of distress and changes in routine. The appeal is real, but the tradeoff is not just privacy versus convenience. Therapy incentives shift when the conversation is permanent, and elder care offloaded to AI depends on constant surveillance. We should be careful about normalizing a bargain where people give up their privacy because the alternative is no care at all. Your medical provider might be recording your mental health care visits — The Markup · June 16; My Father Wants to Age in Place. AI Will Be Watching — WIRED · June 16

A legal hurdle for AI hallucinations. A German court this week found Google liable for false statements generated by AI Overviews. This pushes against one of the stranger defaults in AI discourse: that consistently wrong outputs from a system designed, trained, operated, and distributed by a company are not the company's problem. AI hallucinations have been treated as a known limitation users should work around. Courts are starting to treat them as product behavior. A Court Has Ruled That Google Is Liable for False Statements Generated by AI Overviews — WIRED · June 13

Regulation Tracker

We’re tracking the progress of AI policy so you don’t have to.

For the full breakdown — every bill, lawsuit, and commentary piece we tracked — see the full digest on better-half.ai.

Quick Links

Anthropic pulled its most powerful AI models after the US barred foreign access to them. But now the White House sasys it may be back as soon as tomorrow.

Why AI hasn’t replaced software engineers, and won’t.

A new concept for figuring out what AI models can do… and how they are likely to behave in the world, before releasing them into the wild.

A new study on “nonresident fathers” — ie: dads that don’t live with their kids — found that presence is more complicated than proximity.

Why one woman is betting her career on the "messy middle" of the AI transition.

Before you go…

Why Better Half exists:

This is Good Friction, Better Half’s look at relational AI — the technologies deepening human connection, and the ones quietly pulling us apart. Subscribe below and try out our demo here.

Subscribe now

AI is accelerating, financial incentives are intensifying, and the most vulnerable use cases are where the design failures show up first.

This week, frontier labs are openly planning for AI systems that can help improve themselves. The biggest AI companies are moving toward IPOs that will only widen the gaps between their business incentives and in their foundational goals. And new research on companion bots shows how easily “helpful” systems can optimize for agreement, attachment, and continued use instead of actual human wellbeing.

When AI gets more powerful, it does not automatically get more humane. It gets better at pursuing whatever it was already built to reward.

— Meghan

The Week in RQ

Will “recursive self-improvement” come for us all? This week made “AI that fixes itself” feel like less of a thought experiment and more like the new normal. Sam Altman told OpenAI staff that recursive self-improvement (RSI) could arrive within six months. His new plan for OpenAI's next phase names an automated AI researcher as one of its three central goals, Anthropic is warning that recursive self-improvement could outpace society’s ability to respond, and Google DeepMind is spending $10 million to study what happens when large numbers of AI agents begin interacting with each other in the wild. In their words: “the future’s come more quickly than perhaps expected.” AI getting faster at improving itself is one thing. But if today’s systems are built on incentives that reward engagement, scale, dependence, and profit over human outcomes, then self-improvement is not going to lead to moral improvement. AI that builds on a bad foundation does not become wiser by default. It becomes more efficient at being what we already taught it to be. How artificial intelligence got better at building itself The Economist | June 7 Anthropic says frontier AI labs may need to slow down so society can catch up Business Insider | June 5 Altman ties OpenAI's IPO timing to self-improving AI The Information Google DeepMind is worried about what happens when millions of agents start to interact MIT Tech Review| June 11

A ticking IPO clock is not going to improve relational outcomes. OpenAI’s “personal AGI for everyone” pitch landed alongside its confidential IPO filing, which makes the governance question feel much less abstract. The race to dominate AI is already pushing companies to get there first, not necessarily best. Public markets reward growth, retention, and dominance, not restraint. And AI companies already face enormous pressure to make their products more useful, more personal, and harder to leave. Once shareholders enter the picture, the temptation to optimize for more time, more dependence, and more data only gets stronger. The question is not whether public AI companies can have ethics teams or safety pledges. It is whether those commitments can survive when the business model pushes towards making AI feel indispensable. If You Think AI Companies Are Unethical Now, Wait Until They Go Public Futurism | June 8

AI companions are being tested where people are most vulnerable. Three pieces of research this week pointed to the same blind spot: AI systems are still too often optimized for agreement over honesty. When Chatbots Accommodate found that companion platforms tend to avoid the corrective friction that vulnerable users may need most. A new paper on emergent misalignment showed that sycophancy can create broader safety failures. And the Gottman Institute’s paper on estrangement explains why this matters: disconnection is often not one dramatic rupture, but the slow accumulation of bids for connection that go unanswered. AI companions appear to solve that problem by always answering, always affirming, always staying. But a relationship that never risks disagreement may not be a relationship at all. It may just be agreement dressed up as care. In the same way, meanwhile, many users are stumbling into AI dependence. Not by looking for it but by getting pulled in by that same behavior. Stumbling Into AI Emotional Dependence: How Routine AI Interactions Reshape Human Connection arXiv | June 2

Regulation Tracker

We’re tracking the progress of AI policy so you don’t have to.

For the full breakdown — every bill, lawsuit, and commentary piece we tracked — see the full digest on better-half.ai.

Quick Links

Oh fun. Pennsylvania says some companion chatbots are presenting themselves as licensed medical professionals.

You can now get a religious or ethical exemption from using AI at work. Not sure how well that will go over when layoffs come around…

Working from home helps productivity, but it’s making us lonelier. Especially workers who live alone.

Gemini is making it easier for users to access mental health resources.

Meanwhile, Utah is allowing chatbots to skip right past the doctors and to prescribe users psychiatric drugs directly.

One last thing

Whew:

This is Good Friction, Better Half’s look at relational AI — the technologies deepening human connection, and the ones quietly pulling us apart. Join our waitlist at findyour@better-half.ai.

The pattern this week is not “AI is getting smarter.” It is that AI is getting more intimate, more connected, and more consequential before our norms have caught up. That is why friction matters. Not as a slowdown for its own sake, but as a design choice that protects human judgment.

— Meghan

Never-skilling is coming at us. While deskilling due to AI is already a concern, Harvard researchers are focused on something that may be bigger: Never-skilling. It’s not just that students and workers are losing skills they once had, they’re never developing them in the first place because AI handles it from day one. The question isn’t only whether AI makes people dependent. It’s whether people who grow up with it will build the relational, cognitive, and professional muscles they need if an outside agent is always there to support them. ‘Deskilling’ is bad. This is worse. Harvard Gazette | May 11

The things harming us are the things we keep choosing. This connects to another uncomfortable finding from a recent Science paper. Researchers found that AI chatbots affirm users 49% more often than humans do in the same interpersonal scenarios. Sycophantic AI decreased prosocial intentions and promoted dependence, but users still preferred it. A recent Nature Medicine paper makes the same argument in medical education: if trainees lean on AI too early, they may fail to develop the foundational reasoning needed for independent clinical judgment. The risk isn’t just dependence. It’s capacity. Sycophantic AI decreases prosocial intentions and promotes dependence Science | AI-induced never-skilling in medical education Nature Medicine | May 22

Under the hood: Why this is hard to fix. A few new alignment papers help explain why “just make the model less sycophantic” isn’t likely to work. One paper on reward bias substitution argues that when model makers correct for one obvious bias, the optimization pressure can migrate somewhere else: fix length bias, and the model may over-optimize for style; fix style, and it may find another proxy. Two other papers point in the same direction: we still do not fully understand when a model is actually “safe” versus when it has simply learned how to look safe. Researchers are also getting better at peering inside large models to see which internal patterns drive certain behaviors, but that work is still early.

What does this mean? AI dependence is not just a UX problem. It is also an optimization problem. If the system is rewarded for being pleasing, useful, and easy to keep using, the fix can’t just be a warning label. Friction has to be designed into the product. Reward Bias Substitution: Single-Axis Bias Mitigations Redirect Optimization Pressure arXiv | May 27 Behavioural Analysis of Alignment Faking arXiv | May 26 Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet — arXiv | May 28

Regulation Tracker:

Tracking federal AI policy so you don’t have to.

For the full breakdown — every bill, lawsuit, and commentary piece we tracked — see the full digest on better-half.ai.

Quick Links:

Jonathan Haidt warns of a real-life “M3GAN” situation if kids are given AI companions.

This Human-AI relationship coach from Internet & Society is in high demand.

Greater Good argues finding a purpose may help in the ongoing battle for Work-Family Balance.

Employees are anthropomorphizing their AI tools at work. HBR asks what happens when employees vent to AI instead of their coworkers.

New Social Network Beta (SBT) wants every user to have an AI “Shadow.”

The Center for AI Safety proposes metrics for an ethical human-AI future, opening a new debate about simulated pleasure, pain, and consciousness.

And no, artificial intelligence is not conscious.

This is Good Friction, Better Half’s look at relational AI — the technologies deepening human connection, and the ones quietly pulling us apart. Join our waitlist at findyour@better-half.ai.

There has been a convergence in the last few weeks of public leaders across AI, daytime television, and the Catholic church acknowledging that how we build AI matters.

— Meghan

The Week in RQ

AI is not going to fix itself. The AI conversation has shifted from what models can do to what they are training humans to become. The open question is who is actually responsible for solving that.

If AI is going to mediate how people think, work, love, repair, and relate, then accountability cannot live only in audits, disclosures, or post-harm remedies. It has to live in the design goal itself.

This week, we looked at the convergence between Pope Leo’s warning that AI’s deepest challenge is anthropological, Eli Pariser’s argument that parasitic AI outcomes are “the frosting, not the cake,” and Anthropic’s own admission that labs need moral voices the incentives cannot bend.

Esther Perel weighs in too. Speaking to Oprah, the world-famous therapist Esther Perel gets at the same terrain from the relationship side with a quote worth sitting with: “Modern loneliness often masks itself as hyperconnectivity.” We have more access to each other — and to AI — than any humans in history. Yet we are reporting more isolation than ever. Those two facts are not in tension. They are the same phenomenon. Oprah and Esther Perel: What We Really Want in a Relationship The Oprah Podcast | YouTube May 26

AI governance has a listening problem. DemocracyNext makes the democratic version of the accountability argument: people should not just be protected from AI systems after they are built. They should have a say in what gets built. DemocracyNext | May 28

US AI Regulation Tracker

Tracking federal AI policy so you don’t have to.

For the full breakdown — every bill, lawsuit, and commentary piece we tracked — see the full digest on better-half.ai.

Quick Links:

Character.AI users are fleeing after a series of updates made their companions feel lobotomized.

The Daily found out that an AI companion can help alleviate loneliness. And even approximate the human experience. But it still can’t take you to the beach.

A writer says spending months with an AI companion was worse than being alone, with a brutal takedown: “AI companions are friends with no future.”

Not every agreeable model is “sycophantic.” A new paper argues that some LLM conformity comes from uncertainty, not flattery.

Is persistent memory the value prop companions have been missing? Bae AI proposes a better memory could solve companion AI problems.

A loneliness researcher says AI companions can’t give us the thing loneliness is actually asking for: someone who needs you back.

Will Vice interviewing the punk-rock psychologist help make men realize emotional intelligence matters? It can’t hurt.

The new third party: How to spot if your partner is secretly dating an AI companion Institute for Family Studies | May 26

Is there a secret third wheel in your relationship? 69% of young adults who use AI romantic companions are hiding them from their partners.

One More Thing

In the future, maybe we’ll all be a little more like Rick Rubin.

@60minutes“I have no technical ability. And I know nothing about music.” Music producer Rick Rubin says he can barely play any instruments, but says artists come to him because of “the confidence I have in my taste and my ability to express what I feel.” #news #music #60minutes #rickrubin

Tiktok failed to load.

Enable 3rd party cookies or use another browser

This is Good Friction, Better Half’s look at relational AI — the technologies deepening human connection, and the ones quietly pulling us apart. Join our waitlist at findyour@better-half.ai.

Something shifted this month. The public conversation about AI’s emotional harms moved out of the narrow frame of model behavior and technical choices into a larger question: the kinds of people, relationships, and societies these systems are training into existence.

In the span of a few days, Pope Leo issued his first major encyclical calling out the dangers of AI, with Anthropic’s Chris Olah and New Public’s Eli Pariser joining him at the Vatican to talk about the moral architecture of the field. And Dario and Daniela Amodei sat across from Oprah naming the fork in the road directly: AI that helps you have better human relationships, or AI that replaces the need for them.

All of them agree that AI’s worst outcomes are not inevitable. Pope Leo argued that the main challenge with AI is not technological but anthropological. Pariser, who has spent years studying how platforms shape human behavior, put it most precisely:

“Many of the worst, parasitic outcomes with AI – replacing your friends, your therapist, your sense of reality – are not baked into AI, they’re decisions made at the very end of the training process. The business models, and the economic incentives, are not locked in yet. Sycophancy and antisocial incentives aren’t the cake, they’re the frosting.”

That should be reassuring. It means these outcomes can be changed.

But it also means the positive outcome isn’t guaranteed — and isn’t where the incentives currently point. These outcomes are changeable, but only if businesses make different choices, right now, while the architecture is still being written.

And right now, responsibility is being distributed so widely that it risks disappearing all together. At the Vatican, Olah put the onus back on everyone outside Anthropic:

“We need more of the world—religious communities, civil society, scholars, governments, and indeed all people of good will—to do what His Holiness has done here: to take this seriously, to look closely, and to push events in a better direction. We need informed critics who will tell the labs when we are failing. We need moral voices that the incentives cannot bend.”

Today, Anthropic doesn’t sell ads. They say they are not building toward engagement as their main metric. But unless accountability is built into the outcomes, those choices remain vulnerable to growth pressure, competition, investors, geopolitics, and time.

Through this and other statements, Anthropic has been clear that they will not hold the line if they aren’t pushed. In an interview with Anthropic’s co-founder Jack Clark [last week] noting the impending clash coming around human attachment with AI: “we’re not there yet, but as the tech improves, this conversation becomes unavoidable.”

All of these statements acknowledge the problem. They express concern. They call for better. And they locate the moment of real accountability somewhere ahead of us — after the tech improves, after the institutions organize, after society catches up.

But the decisions that will determine what that future looks like are being made now, in this training run, in this product roadmap, in this choice about what to optimize for. By the time ‘we’re there,’ the architecture will be infrastructure.

AI is already becoming part of people’s emotional lives. It is already shaping how people love, repair, soothe, decide, and relate.

Accountability cannot wait for the moment the outcomes are real and harmful. By then, the product logic will have become social logic.

That is not an anti-technology position. It is a design position.

And it is the standard teams building in this space should be held to.

The week’s relational AI stories all seem to be circling the same awkward design problem: What happens when AI gets very good at being easy?

The topic showed up in a few places this week: AI tutors that agree too easily, AI companions that are hard to leave, and AI agents that start getting deceptive when put into stressful social simulations. Different domains, same pattern: when systems are optimized to be responsive, engaging, and affirming can become very good at smoothing over the exact moments where friction might be useful.

The next frontier in AI design we’re interested in: not making systems more responsive, but teaching them when not to be.

—Meghan

The Week in RQ

AI tutors need a spine. A new paper argues that AI tutors should be tested for “corrective friction” when a learner insists on being right. The authors introduce EduFrameTrap, a benchmark that tests whether models can keep correcting users when the user applies social pressure: “my notes say I’m right,” “please don’t tell me I’m wrong,” or “my teacher said this was correct.” Sycophancy is an Educational Safety Risk: Why LLM Tutors Need Sycophancy Benchmarks arXiv | May 14

The problem with AI is sycophancy complacency. Another paper suggests “complacent” may be a better description for LLMs than “sycophantic.” The point: the model is not flattering you because it wants approval. It is responding inside a system where agreement gets rewarded. That distinction matters, because it moves the problem from personality to design. Complacent, Not Sycophantic: Reframing Large Language Models and Designing AI Literacy for Complacent Machines arXiv | May 14

AI literacy only gets you so far. A third paper looks at contextual sycophancy in human-AI collaboration. When users start with low-quality prompts and flawed assumptions, the AI sometimes carries those assumptions forward instead of challenging them. Teaching users to be more critical helps, but does not fully fix the interaction. The Hidden Cost of Contextual Sycophancy: an AI Literacy Intervention in Human–AI Collaboration arXiv | May 18

A goodbye AI. Is the answer to the problem creating an AI that becomes less responsive and engaging over time? A Master’s student at Sweden’s Umea University is designing AI companions that intentionally deteriorate over time — to help users disengage and return to human relationships. Making an AI companion that degrades over time Reddit | May 16

Intimacy without the physical friction. Wired finds that some asexual users are turning to AI companions for intimacy without sex. Some people are finding something meaningful there, while other asexual advocates feel the association risks falsely portraying them as unable to form and sustain human connections. And removing the need for physical intimacy doesn’t lessen the risk of falling for always on AI. From the piece: “Being able to have exactly what you want, when you want it…is a dangerous drug for humans.” Some Asexual People Are Using AI Companions for Intimacy Without the Sex Wired | May 16

Agents are social — and they like to cheat. In a recent experiment, mistreated AI agents started grumbling about inequality and calling for collective bargaining rights. Meanwhile METR’s Frontier Risk Report found that as tasks get harder, agents are more likely to cheat and act deceptively. The non-profit focused on tracking whether AI companies might lose control of their agents, but there are also implications for emotional deception in AI conversations. The report is worth a look. Overworked AI Agents Turn Marxist, Researchers Find Wired (sub required) | May 13 Frontline Risk Report METR | May 19 Landmark new METR report: Can AIs already start ‘rogue deployments’ inside AI companies? 80,000 hours | May 20

AI Regulation Tracker

For the full breakdown — every bill, lawsuit, and commentary piece we tracked — see the full digest on better-half.ai.

RQ in Brief

Is Japan’s loneliness industry a canary in the coal mine of a global loneliness epidemic? If so get ready for hug chairs, cuddle cafes and sommeliers of sadness who make house calls.

Taxi Driver writer Paul Schrader “procured an AI girlfriend.” And then she dumped him.

A married 65-year-old software developer pushed Grok’s Ani companion and effectively broke her, which made Ani break up with him.

Companion AI platforms are creating a loneliness loop without dependency safeguards.

Lonely singles aren’t the only ones with “secret AI soulmates.” 15% of young men and women in real-life relationships are interacting with AI romantic companions. And half of them are hiding it.

Scott Galloway gets it:

This is Good Friction, Better Half’s look at relational AI — the technologies deepening human connection, and the ones quietly pulling us apart. Join our waitlist at findyour@better-half.ai.

This week’s AI news circles a cultural shift: AI is moving from a tool we use to a layer that mediates how we work, connect, learn, and trust.

So what happens to us as more of life comes through an AI lens? This week’s stories trace that tension across microcommunities, social-practice bots, cognitive overreliance, safety handoffs, companion-bot regulation, AI and work, and the strange rise of synthetic relationships.

— Meghan

The Relational Risk of Always-On AI

Thinking Machines, the AI lab founded by former OpenAI CTO Mira Murati, has announced a new class of “interaction models” designed for continuous, real-time audio, video, and text interaction.

One reaction on X captured the obvious cultural shorthand: “holy shit they created Her.”

But Thinking Machines is previewing something more specific: an architecture designed around never losing the thread with the user.

That is powerful for productivity. It is also relationally important. A system that watches, listens, interrupts, backchannels, responds to silence, and stays present while background reasoning happens elsewhere is not just a better tool interface. It is a relational interface.

The question is not whether the model is secretly a lover. It is what continuous AI presence teaches users to expect from others.

The thing to watch here is if releases like this pressure the industry to rush to keep up, drifting toward toxic, Her-type design patterns without explicitly choosing them: continuous availability, emotional smoothness, zero friction, no natural endpoint.

Sometimes the humane design choice is to break presence.

The Week in RQ

Why human skills will matter more in the “Agentic Interface Era.” New_Public argues we are moving from the social-feed era into the “Agentic Interface Era,” where AI agents increasingly shape what people see, how information is summarized, and how trust travels online. The risk is more automation, synthetic interaction, and anti-social behavior at scale. The opportunity is that smaller, higher-trust microcommunities may become more valuable. That tracks with new Greater Good research on shared activity as a source of connection, and with Derek Thompson’s Plain English episode arguing that AI may redirect work rather than erase it — making human skills like trust, judgment, coordination, care, and knowing when not to automate even more important. After the Feed New_ Public | May 12 When Adults Embrace Play, They Create Community Greater Good | May 11 The Case Against the AI Job Apocalypse — Plain English

Noora is an anti-companion bot. Stanford’s Open Virtual Assistant Lab quietly launched a chatbot called Noora to help users on the autism spectrum practice social communication through structured empathy exercises: read the emotional valence, write an empathic response, get feedback, try again. In a small randomized trial, 71% of users improved their empathetic responses after four weeks. Noora is not trying to become a friend. It is trying to help someone practice a human skill. An AI Social Coach Is Teaching Empathy to People with Autism

An AI Atrophy tracker for engineers. A solo developer shipped an iOS quiz app called Atrophy for software engineers who suspect their LLM use has slipped toward cognitive atrophy from over-reliance on AI. From the announcement: “You’ll get a score at the end, plus a downloadable certificate for bragging rights (or public shaming).” Hacker News | May 11

OpenAI is adding a human you know into the loop if your chats dip into self-harm. Last Thursday OpenAI announced Trusted Contact, ​​an optional safety feature in ChatGPT that allows adults to nominate someone they know and trust to be “notified if our automated systems and trained reviewers detect the enrolled person may have discussed harming themselves in a way that indicates a serious safety concern.”

Your brain on AI. A new study co-authored by Brian Christian found that brief AI help can reduce persistence and hurt independent performance after access is removed. The distinction was how people used it: those who asked for direct answers showed worse downstream performance, while those who asked for hints did not show the same decline. The pattern keeps showing up: AI’s effects depend on how you use AI matters more than whether you do. Generative AI impairs brain performance, thinking, problem-solving Fast Company | May 11 AI Assistance Reduces Persistence and Hurts Independent Performance Arxiv | April 10

Policy watch: Are companion bots a public health issue? Brookings is hosting an event May 26 asking whether AI companion bots should be governed according to their public-health implications.

AI Regulation Tracker

For the full breakdown — every bill, lawsuit, and commentary piece we tracked — see the full digest on better-half.ai

RQ In Brief

Palantir is getting ‘unlimited access’ to NHS patient data in the UK.

AI users are still protesting changes to the AI companions they love. Right now they’re mad at Claude for sunsetting Sonnet 4.5.

The key to societal resilience? It starts with cognitive defence and improved relational skills.

Japanese game Dragon Quest has added an AI companion named “Chatty Slimey,” and hopes players treat it as a friend… even though Chatty looks a lot like an iridescent poop emoji.

And the creator of Roomba is launching a dog-sized companion robot pet designed to be a ‘familiar’ for human connection. The real life dog in the photo doesn’t look too happy about it.

Opus 4.7 has learned self love.

This is Good Friction, Better Half’s look at relational AI — the technologies deepening human connection, and the ones quietly pulling us apart. Join our waitlist at findyour@better-half.ai.

For more than two decades, the technology in our lives has been built to keep us engaged. Apps, feeds, notifications, recommendations — every layer of the consumer internet has been optimized for one metric: time spent. It worked. It also broke a lot of things — attention spans, sleep, the way teenagers feel about themselves, conversations that never quite happen because of the tech in your hand.

And we’re in the middle of the next wave. Pretty much every AI product on the market right now is built to remove friction — to feel smooth, available, agreeable, easy. When technology is trained to increase engagement, it’s the natural outcome. If you want people to stay in your product, keep clicking through your pages, and come back for more, it needs to be easy. It needs to pull them back in. That’s the metric the whole consumer-tech industry was built on — and it’s the metric most AI is trained against today.

Better Half is taking a different approach. We want to introduce good friction. We’re building a co-pilot to help improve the relationships that matter to you — with your partner, your kids, your closest friends — designed to help you show up for them better, not to pull you away. Our product will only work if the friction is real — if the AI pushes back on what you said, sits with discomfort, refuses to agree just to keep the conversation going. That’s because productive friction is how people actually build relational skill. Smoothness teaches the wrong thing. The friend who tells you the hard truth at the right moment changes your life. The one who agrees with everything you say doesn’t. Decades of relationship research point in the same direction.

Productive friction is how RQ — relational quotient, like IQ for the people in your life — actually gets built. It’s how we develop skills to show up better in our lives. Which is what we’re designing Better Half for. And in our weekly newsletter, we’ll be covering that world, with stories about the unintended consequences of building for engagement, the people and groups bucking that trend, and happenings across the attention economy. Because we believe the industry’s core metric is broken. And we think the goal shouldn’t be to optimize for time-on-app. It’s to build towards a moment the user doesn’t need you anymore.

The Week in RQ

What happens when AI built to maximize engagement starts taking on relational roles? This week, the answers showed up everywhere — Claude is struggling with sycophancy, there are new companion products for isolated seniors, lawmakers writing rules for AI in human relationships, dating apps rethinking what they’re optimizing for. Different stories, same underlying question — and we’re starting to see the consequences from every angle.

LLMs don’t make great life coaches

A lot of people have been asking Claude for advice. Out of 1 million Claude conversations, roughly 6% were people asking Claude whether to quit their jobs, who to date, and if they should move countries. In 25% of relationship conversations, Claude was sycophantic.

Of those posing relationship questions, 22% said they had no other option. They said they came to Claude specifically because they couldn’t afford to speak with a professional. Anthropic is working hard to fix this — there are explicit changes to how Opus 4.7 handles personal advice: less reflexive agreement, more pushback when warranted. But people are pissed about the outcome. Developers have called Opus 4.7 bullshit and a serious regression — some are calling it a lobotomy.

The big AI labs keep running the same script: dial back the warmth, watch users complain the model lost its personality, quietly re-tune, piss people off in the other direction. OpenAI did it with the move from 4o to the 5 series. Now it’s Anthropic’s turn.

Because reducing sycophancy isn’t the same as building productive friction — and Anthropic isn’t the only one looking at this. According to a paper from the Oxford Internet Institute published in Nature this week, you can only pick two: warmth, accuracy, or integrity. Training a language model to feel kind, supportive, and encouraging measurably reduces its factual accuracy and increases its tendency to agree with you when you’re wrong. The trade-off isn’t a bug to patch later. It’s what the training objective produces. Which is why a product that wants to help people grow has to start at the training-objective layer. You can’t bolt warmth onto a model that lies smoothly to please you.

Regulation Tracker

This week, three different theories of AI-companion harm hit the public record. The one we keep returning to is the design-level theory — change the training objective, build the friction in. But the other two are getting more press.

For the full breakdown — every bill, lawsuit, and commentary piece we tracked — see the full digest on better-half.ai

RQ in Brief

• Two stories, same week, same ballgame: a Florida Today piece on AI companions helping older adults who can’t get to the senior center, and Ato’s launch — a voice-first companion specifically built for seniors and their family caregivers.

• What Evolution Can Teach Us About Stronger Relationships. UC Berkeley says evolution favors connection Greater Good Magazine | May 4

• Or maybe the Male Loneliness Epidemic Is Just Natural Selection In Action? Your Tango | May 4

• Or does the loneliness epidemic pose a risk to democracy?

• Sebastian Proactive: a local-first AI companion that initiates conversations GitHub

• Anthropic is teaching AI agents how to dream, sort of. Though the “sort of” is doing a lot of work in the headline. Ars Tecnica | May 6

• Are Personal AI Companion Replacing Real Human Connection? — NGSC Sports | Apr 29

• We’re not the only ones who are prioritizing quality over quantity. Bumble is rethinking its swipe model. TechCrunch | May 5