We Shape Our Tools, and Then They Shape Us

I’m the founder and CEO of Better Half, a relational AI that slashes sycophancy and actively trains the hard, counterintuitive, and often personally confronting social skills people need to improve their relationships and deepen their capacity for authentic connection.

The how is our proprietary decision engine, but I won’t bore you with that. Because the why is a lot more interesting.

Today, I’d like to take you behind the scenes a little bit.

The road to Better Half began when I YOLO’d back to grad school in 2020 (M.S. in privacy and infosec at UT Austin) — and my thesis named a category I thought needed to exist: “tech for human agency.” It caught on more than I expected. A venture fund eventually adopted it as its investment thesis, and for the past few years I’ve been collaborating with them to shift what gets built and funded. (More on that here if you’re curious.)

That front-row seat to founders and venture made one thing clear: human agency doesn’t go far enough. There’s a missing piece, and it’s relational intelligence — RQ.

You’ve seen the headlines: lawsuits over chatbots, regulators fining companion apps, rising reports of dependency, “LLM psychosis,” and a new household term — “sycophancy” — previously a word only shrewd political operatives and admirers of Niccolò Machiavelli would throw into casual conversation. Even the Vatican weighed in — and they don’t liveblog current events casually!

Different stories, same pattern: technology built to maximize retention and engagement predictably drives isolation, erodes our autonomy, and leaves us more fragile and easier to manipulate.

The same logic applies to AI. A model trained to drive loyalty and complete tasks efficiently does fine for code, projects, and email, but it too often fails the moment a person turns to it for relationship help or emotional support.

Which, outside Silicon Valley, is what most people actually use AI for.

They’re about building trust. About navigating disagreement without retreating into avoidance. About repairing after rupture — and learning to hold perspectives that differ from our own.

Those goals aren’t well represented in the objective function for most leading foundation models. Which is why AI gets it wrong so often when it comes to human relationships.

And importantly, the training data for what healthy human relating actually looks like? That doesn’t exist online, so it doesn’t make it into the corpus AI trains on.

Think about what AI learns from: books and essays (passive knowledge, not relating in action), code, the museum of dysfunction that is social media, and therapy transcripts that overindex on pathology.

So AI has no idea how to be with us in the moments that matter most — let alone help us find our way back to each other. Even if we adjusted the objective function, the training data would still be missing.

Which brings us back to the same problem: 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 thereafter our tools shape us.”
— Marshall McLuhan

The tools we build invariably train us back. They build habits we're not even consciously aware of. And how we interact with AI quietly shapes how we show up with the humans in our lives. Because we don’t easily segment who we talk to — human or AI. We learn from AI that habitually validates our limited perspectives, is always available to serve our every whim, and rarely refuses a request that would violate most people’s most basic boundaries.

This invisible behavior modification accumulates one person at a time. And when we act on those learned habits, we shape the relationships that build the social fabric around us — and create the world we inherit every day.

The longer we wait to start shaping our tools in the direction of human thriving, the harder it gets to wrest our agency back and build a future we actually want to live in — not just live with.

Right now, the “default settings” in tech often make an already frazzled, polarized, algorithmically optimized society even more disconnected.

But defaults can be changed. There’s no law of nature that deterministically states that if we build AI, it will go badly. Everything is a function of incentives and system design.

“Show me the incentive and I’ll show you the outcome.”
— Charlie Munger

Better Half is a tool built to deepen human connection by training the skills resilient relationships require — things like conflict navigation, emotional triage, coregulation, and repair after a painful rupture. We draw on time-tested wisdom traditions for direction.

And we don’t track retention or time-on-app — so there’s no reward for drifting into sycophancy or validation when some tough love from a “droid” that never pretends to be human would serve the user better.

We only care about real-life outcomes that lead to social flourishing.

If you agree with the direction we’re taking at Better Half, this is your invitation to reach out directly. Drop a comment, share this Substack, or help us test the product — our MVP is out, and we’d love your feedback while we’re still early.