Catching Yourself in a Riptide
On the quiet undertow of AI, and how to swim out.
The me from last year would be surprised by the ways I’ve used AI in 2026.
Claude has wordsmithed holiday toasts. It’s helped me analyze my personal finances, and therefore now knows exactly where my money is stashed. It helped me navigate an adoption process for my daughter. And it’s been the next stop to process a therapy appointment...you know, for a second opinion.
A big part of my job at the moment is helping people navigate using AI intentionally. And still, every one of those examples represents an insertion into a part of my life I would have imagined I’d guard, or a disclosure of sensitive information that the me of yesterday would have thought was a tad reckless.
How did I get here?
To be fair, I’ve decidedly opted into not being a luddite on this particular wave. I was one to dip my toe in early to see what AI could offer me. And while not universal, the willingness to wade in has proven eye-opening and incredibly helpful in running a small company, in being a new dad, in trying to stay on top of the many ways the world is shifting.
But the bargain implicit in exploring this tech is that you give it more, it gives you more. Layer that on top of the ambient feeling that everyone else already figured this out. Wait, it can do that? How do I stay on the cutting edge?
Growing up, my family and I would take trips to the Outer Banks. Even in a family of competent swimmers, we were always being warned about riptides. You can be staring out at a very normal-seeming ocean, they would warn. You are swimming comfortably near the shore and then lift your head up to find yourself in deep water. It seemed apocryphal until that summer when we watched my incredibly strong uncle bopping up a hundred yards offshore.
Lifting my head above water right now and taking stock of the relationship I have with AI, it feels the same. Some of these shifts like using AI to parent, or mistrusting human advice in favor of algorithmic guidance, happened by nature of simply discovering what was possible. And even when seen with clear eyes today, a lot feel worth it.
But what is disconcerting is the quiet undertow of these changes that rarely felt like decisions at all. There is an unconsciousness with which you can slide into very different behavioral dynamics that rapidly feel normal. And with that, likely some real ways in which this is rewiring how I think, what I go to people for, and how I see myself.
We see this in our research at The Rithm Project. Many of the young people we interviewed who ended up using AI for emotional support or deeply personal reasons started by leaning on it for homework help or to play a game. They find a lot of those use cases to be exciting and helpful, but in the same breath report how using it has planted seeds of doubt in their personal relationships (”now I wonder if those texts are really from them or AI”).
At The Rithm Project, we’ve made a real bet on agency. We’re counting on people, including young people, to navigate this intentionally. That said, navigating intentionally is still hard.
So we went to the research looking for proven ways to shape behavior against a fast-moving force. Said differently, here are three ways to riptide-proof yourself.
1. Prepare before you swim.
The first rule of rip current safety is simple: know the risks before you get in the water and make a plan.
Research on precommitment (from economist Thomas Schelling and psychologist Dan Ariely, among others) shows something that anyone who’s made a fresh commitment to healthy eating likely knows: willpower fails in the moment.
Precommitment encourages you to set boundaries and conditions for success before a tough decision arrives. Instead of evaluating each time whether AI should help with this, you decide once: “these are my boundaries.”
The move: Pick a shortlist of “rules” you want to hold to in your AI use. I’ve drawn a hard line at personally writing anything destined for loved ones. For you it might be explicitly telling others when you use AI, or also committing to checking in with a human when you’re using AI to process something hard. They shouldn’t be arbitrary. Take stock of the things that you currently do that most help you feel connected to the people around you, and keep those for yourself.
A little help:
If you’re anything like me, this can feel enticing but also daunting to summon up catchy and high-impact boundaries out of nowhere. So we crowdsourced a list of “rules” people in our network are putting around their AI use. Take a skim and pick three to five that are speaking to you, or use them to spark other ideas.
2. Keep an eye on the shore.
Once you’re in the water, the most important thing is to keep checking your position. How far out are you? You check regularly, not because you’re panicking, but because the current is subtle and the drift happens when you’re not looking.
Donald Schön, one of the most influential thinkers on how professionals learn, distinguished between two kinds of reflection. Reflection-on-action is what you do after the fact, looking back at how things went. Reflection-in-action, which is much harder, is noticing what you’re doing while you’re doing it.
Most of the conversation about AI boundaries lives in the reflection-on-action space. You read an article (like this one), feel motivated, maybe make a rule. But the slide doesn’t happen in big moments where you stop and evaluate. It happens in the hundreds of small moments where you’re just trying to get something done and the AI is right there.
The weekly practice below is the easier version: reflection-on-action. Looking back at the week and asking honest questions. The longer game is building the awareness to catch yourself in real time, to notice the moment you’re handing something to AI that you should keep for yourself. That takes longer. Start with the easy version, but make it regular.
The move: Once a week, two questions:
How did I use AI to help strengthen my relationships this week?
Where did I use AI in ways that might have pulled me from people, even in subtle ways?
You can do this alone, but accountability helps. Create a standing 15-minute check-in with a friend or colleague. Or schedule your AI to actually help you by setting a weekly prompt that s`ources data points from your recent conversations. This past week, my prompt surfaced a good one: I’d looked up 3-year-old developmental milestones and turned them into a conversation starter with my husband to make meaning together. On the flipside, it flagged that going straight from therapy to Claude for further analysis might have been a moment to sit with it myself, or take it to someone I trust.
A little help:
Here’s a prompt you can copy into Claude or ChatGPT to set up a weekly reflection that runs automatically.
Weekly reflection (paste into your AI memory/instructions):
Every Friday, review my recent conversations with you from the past week and ask me two questions:
How did I use AI to help strengthen my relationships this week?
Where did I use AI in ways that might have pulled me from people, even in subtle ways?
Before asking, surface two or three specific moments from our recent conversations that might be relevant. Don’t judge them. Just name them so I have something concrete to reflect on.
After I respond, don’t try to fix anything or offer advice. Just help me notice the pattern.
3. Swim in safe water.
You can’t eliminate rip currents. But you can choose where you get in, pick safer beaches, stay close to lifeguards. You can make the water you’re in safer.
The first two moves are about awareness and willpower. Decide in advance. Reflect each week. Both are worth doing. But anyone who’s ever set a New Year’s resolution knows that awareness and willpower only carry you so far.
Chip and Dan Heath have a useful metaphor for this in their book Switch. They describe behavior change as a rider on an elephant walking down a path. The rider is your rational mind, the part that makes plans and sets rules. The elephant is your emotional side, which provides energy but follows the path of least resistance. The path is the environment you’re operating in, and when the path makes the right behavior easy, the rider and the elephant don’t have to work so hard.
The beauty of most AI platforms is that they are steerable, to some degree. Already we see examples of young people hacking the tools to mitigate their riskiest tendencies. Just a few of the prompts we’ve heard recently:
Assign a role that changes the register. “I tell it to respond like a sassy aunt who loves you but can’t help but tell you how it is.”
Ask for multiple perspectives, not one. “How do you think the other person would think about this?”
Borrow a specific expertise. “Channel Esther Perel on this relationship challenge.”
Name what you don’t want it to do. “Don’t just validate me.” “Don’t rush to fix.” “Push back if I’m being avoidant.”
You can do this in one-off ways, but also hardcode the tools you use most often to make it less likely they will lean into the riskiest social behaviors, or at least add some healthy friction to make those choices more obvious.
The move: Jason Weeby, a strategist and close collaborator of The Rithm Project, helps organizations effectively adopt AI. In ensuring he’s on the cutting edge, he has also been thoughtful about the benefits AI offers and the risks that come with them. Tracking the threats of cognitive decay that can come with outsourcing to and collaborating with AI, he wrote a great piece: AI Will Make Me Dumber If I Let It. So I Made a Helmet. That helmet? A set of concrete, research-backed directives that he added to Claude’s memory so that every time it responds to him or acts on his behalf, it honors those principles. He did the work to understand those intentions (and even printed out a set of them to keep them present, similar to the rules in #1). But he didn’t rely solely on that. He changed the architecture of the tool itself.
Jason’s helmet is focused on cognition, on how AI affects the way you think. The risks we’re watching most closely at Rithm are relational. How AI reshapes the way you show up with other people, what happens to your tolerance for real human friction after weeks of frictionless AI, and whose information you’re sharing without thinking to ask.
Below is starter language for what a relational version might look like. Paste it into your AI’s memory or custom instructions, try it for a few weeks, and revise based on what you notice.
A little help:
Relational guardrails (adjust/edit and paste into your AI memory/instructions):
When I ask you to help write a message to someone close to me, give me back my own words with light edits for clarity. Do not rewrite my voice or upgrade my warmth. If my draft is messy, that’s fine. The person receiving it is in a relationship with me, not with you.
When I bring you a conflict or frustration with someone in my life, ask me first: “Have you talked to them about this?” If I haven’t, help me figure out what I want to say to them directly rather than helping me process the situation in the abstract.
If I share someone else’s words, messages, or personal information with you, ask me: “Do they know you’re sharing this?” I may have a good reason. But I want to be asked.
Do not let me rehearse a difficult conversation with you more than once without asking whether I’m ready to have the real one.
If I seem to be using you as a substitute for sitting with discomfort, name it. Say: “This might be worth sitting with before we work through it.”
Once a week, ask me: “Where did I show up instead of you this week?”
👉 Read Jason’s full approach: AI Will Make Me Dumber If I Let It. So I Made a Helmet.
Swimming Parallel
These strategies aren’t meant to be permanent. They’re meant to help you navigate the currents intentionally and notice the difference between surfing a wave and being drawn out by a riptide.
That’s easier said than done for something changing by the minute, infiltrating new parts of your life, and surrounded by a cultural swirl with no semblance of cultural consensus. It’s unlikely to just happen.
We see this in young people, too. In our research, the ones who navigate AI best aren’t the abstainers. They’re the ones who are taking an active stance in defining their relationship with AI, in seeing it as something they can morph and mold, and perhaps most importantly, for having real humans in their life they can be honest with about this process.
In an ocean, if you find yourself in a riptide, you’re not supposed to fight against it. You swim perpendicular, breaking the pull just enough to bring consciousness and freedom back to where you want to go.
Safe swimming.
This is a great reminder to be mindful of how we are using AI, including before and during. Important practices. We used to vacation at OBX, too! Riptides were something I learned about from a young age, and I really like the metaphor as used here.
As a designer, artist, and critical thinker, I found myself distracted in reading this post, wondering whether or not the illustrations were created by AI? It's not just words and text messages that bring the uncertain feelings and doubt.