Complementary, Not Interchangeable

A friend emailed me last week.

“Sara. I just finished reading your snail mail which then sent me to your musings on AI. Jana shared this essay with me & I thought you may enjoy reading it. Hope all is well. 💞xo”

(I love that this is still how things reach me sometimes. Not an algorithm. A person who read something, thought of me, and went digging through her own inbox to find it.)

I was on the couch when I opened it, working through the non-work messages I’d been ignoring all day, about ten minutes out from my actual evening, which is a fizzy water and music and then either Spanish practice or a real book with pages. I don’t always finish what people send me to read. I get the point halfway down and I’m already onto the next thing. But this one I read all the way through. Then I just sat there for a minute, because it said something I’ve been circling for months without landing on the language.

The essay was by a writer named Eileen Workman, and it was about AI. She wrote it in cooperation with AI, and she points that out for good reason. The dialogue is also the demonstration. Both of their inputs in the same piece.

She starts with a question I’ve been circling too. Why does almost every conversation we have about AI start from the same assumption, that eventually either humans will control AI or AI will control us? One or the other, pick your ending. I think we just want to know how the story turns out so it feels less uncertain, and preferably we’d like it to be one of the classic happy stories. But life doesn’t ever really work in that tidy sort of way (outside of movies).

Eileen suggests that this “who’s in charge” assumption tells us less about intelligence than it tells us about ourselves, because we have spent thousands of years organizing everything into hierarchies. Ruler and ruled. Owner and owned. We talk about nature that way too, gaining mastery over it, like it’s something to be won. It’s simple, and it gives someone a lot of power in every scenario, so of course it’s appealing.

When something new, smart and exceptionally powerful shows up, we go looking for the hierarchy. Where is our place? Hopefully on top.

What if neither of us has to be on top of the other? What if that isn’t the only option here?

Do you ever notice how hard it is to even hold that question? Reciprocity, equity, balanced power, it almost feels like some utopian concept. Maybe it is. But I’m hanging on to the hope that it’s a real possibility.

The question of power stopped feeling philosophical to me this month, and it’s because of an app.

Meta launched an app called Muse on September 8th. Within about two weeks it was the top free app on the US iPhone App Store, ahead of ChatGPT, ahead of Claude, ahead of Meta’s own AI app, with something like two and a half million downloads. Which, fine, app store charts move around. That’s not the part that got my attention.

Muse isn’t a chatbot. It’s an agent. You connect it to your email, your calendar, your payment accounts, your messages, and then you tell it what you want done and it goes and does it. Opens browsers. Fills out forms. Books the travel.

So the thing I’ve been saying for a year, the question I ask myself about every single AI tool, what am I handing over and what is it handing back, just stopped being a metaphor. People are handing over actual account access to actual money. And they’re doing it by the million, in two weeks, mostly without reading anything about it, because the basic version is free and the pitch is that your life gets easier.

I’m not going to tell you that’s bad. I don’t know that it’s bad. (I do know I’m not doing it yet, and I know why, which I’ll get to.) What I’m saying is that we crossed a line this month from “AI helps me think” to “AI acts for me,” and almost nobody stopped to decide anything about it. It just happened, the way things do now.

Which brings me back to Eileen, and the thing I’ve been circling.

She points out that she brings things to her AI that it doesn’t have. The human experience. A body. Pleasure and pain. People she loves and worries about. Memories that mean something because she actually lived them. Values she picked before she ever asked a machine for anything.

And it brings things she doesn’t have. It reads more than she could in a lifetime. It notices connections. It challenges her reasoning without getting tired or offended or huffy about the fourth version of the same question.

Eileen and her AI are not the same thing. We know that logically. But it takes a closer look to see that the difference between them is the whole point.

There’s a research term for this, human-AI complementarity, which is a clunky phrase for a simple idea: stop asking who does it better, start asking what the two of us together can do that neither of us can do alone.

The clearest example I have is my website.

I know what a good website looks like. I can look at a page and tell you the spacing is off, the text isn’t written for our audience, that button should be somewhere else, this whole section feels cold. I have strong, specific, useful opinions about design. What I cannot do, and will never do, and honestly might be the worst possible job for my dyslexic brain, is write the code. So I give feedback and Claude writes the code and I look at it and give more feedback, and we go back and forth like that until the page is right. Neither of us could have made that page alone. I couldn’t write it. Claude doesn’t know what I want.

Do you see what I mean though? That’s not a tool. A hammer doesn’t ask clarifying questions. But it’s also not a partnership of equals, because I’m the only one in the conversation who truly cares how it turns out.

It’s something that isn’t defined by who holds the most power, or who’s the smartest or funniest or most human (or least human). We don’t have a good word for it yet, which might be why we keep reaching for who’s in charge.

And this is where I think we have something to offer, those of us who do this work (therapists, healers, helpers, communicators, artists, creatives...).

Eileen says that a genuinely cooperative relationship would need boundaries. Autonomy. Consent. Privacy. The ability to refuse, which she calls sovereignty. She says no intelligence worth trusting should require us to surrender ourselves in order to get its benefits.

I read that and thought, oh. That’s what we do every day. It’s often our whole job.

We spend our careers helping people figure out where they end and someone else begins. We know that boundaries aren’t distance and they aren’t distrust. They’re the thing that makes closeness possible without one person dissolving into the other. We know that consent has to be ongoing and specific, not a one-time yes (or no). We know that “I’ll do this part and not that part” is a complete sentence.

(I want to be careful here. I’m not saying AI is a client or a person or that it has feelings I need to consider. We don’t know what it is yet. I’m saying the skill we already have, for figuring out what to give and what to keep, is exactly the skill this moment requires, and hardly anyone in the AI conversation is bringing it.)

So here’s mine. My actual lines, which took a while to find and have moved before. And will move again. This is all very fast-moving.

I don’t hand over the keys. I’ve given Claude its own door (a small one, with a doorbell), built one task at a time, and I approve what comes through it. Muse, as far as I can tell, is more of a here’s-my-whole-key-ring situation.

I don’t give it my thinking. This is the big one. I almost never ask it for ideas. I ask it to interview me, to ask me harder questions, to push on what I just said, to find the hole in my logic, so that I can dig further into my own thoughts and get the thing that’s in there out where I can see it. AI is my fellow spelunker in the caves of my own mind. My frameworks, my courses, my curriculum, my guides, those come from me. Every time. But I’ll absolutely ask it for forty questions to help me find what’s buried down there.

My voice is mine and I hold onto it with both hands. I read every single word of anything I’ve had help drafting. Every word. Twenty years of writing delicate, deliberate emails has made me pretty good at them, and most of the time I just write them myself because it’s faster. But when I’m behind and I need to send something kind that sounds like me, I’ll ask for help, and then I’ll edit it, and I still read every word before it goes out.

Numbers, though. I use AI for numbers like my life depends on it, and I say that as a person with dyslexia who has spent decades feeling stupid in front of a spreadsheet. Same with grammar and sentence structure, which have never once made sense to me. That help is not a shortcut, it’s access. I get to bring what I’m actually good at instead of drowning in the part that was never going to work.

I don’t use it for clinical approaches or interventions. I might have, if it had existed when I was a new clinician hungry for anything that would get the creative juices going or just some reassurance that I was on the right track, and I don’t judge a new clinician for doing it now. But that’s not where I get my clinical thinking. Notes are different, and I want to be precise about this because I know what you’re thinking: our documentation tool is built for healthcare, with a BAA in place, and a note is a summary of something that already happened in a room. That’s not the same as handing a session to a chatbot. Please don’t do that.

I don’t ask it for recipes. This one has nothing to do with capability. I have cookbooks I love, and there are food bloggers out there who tested that recipe eight times and photographed it and wrote up the note about why you should let the dough rest, and I want them to still be in business next year. Or I make something up myself and it’s great (or it isn’t).

Reading is the same. I want to hear a person’s voice. I don’t want the summary, I want the way they said it. (My voice matters to me, so other people’s voices matter to me. Feels like it should go both ways.)

Images I’ll do, sort of. If I need something generic for a slide, I’ll generate it, the same way I’ve been pulling stock from Canva for years, because that image isn’t art, it’s a visual aid doing a job. When I want real photographs of real life, I hire a photographer. And I make art for art’s sake, which is a completely different thing and has nothing to do with any of this.

And the hard conversations, the condolence note, the message to someone whose parent just died, the check-in with a clinician who’s struggling. Those are mine. If those words don’t come from me they’re not worth sending.

That list is not a rulebook. It’s mine and it’s current and some of it will probably change.

But I notice that having it makes me less anxious, not more. I’m not standing at the edge trying to decide whether this whole thing is good or bad. I decided what I’m giving and what I’m keeping, and now I can just work.

We can ask a better question than who ends up on top. Eileen’s version is what might become possible if neither of us has to lose. Mine is smaller and more Tuesday-morning than that. What am I handing over, what is it handing back, and is that trade one I’d make again tomorrow?

I don’t know how this turns out. Nobody does, including the people who sound certain.

But an essay reached me last week because a woman read something, thought of a friend, and forwarded it. Then that friend read it, thought of me, and dug it back up out of her inbox. Four people and an idea, moving at the speed of people paying attention to each other.

Connection is still the most powerful energy we have to share. That part is still ours.

Referenced: Eileen Workman, “Relationship with AI?”, Facebook, September 14, 2026. Muse figures via CNBC and Sensor Tower, September 2026.