The first merger was a hostile takeover that became a marriage. Two billion years ago, a free-living bacterium got eaten by a larger cell, refused to be digested, and instead settled in as a permanent guest. The deal it cut was outrageous. In exchange for shelter, it would produce all the energy. That bacterium is sitting in every cell of your body right now. We call it a mitochondrion. The merger paid for everything that came after. Bigger cells. Tissues. Nervous systems. Conversations like this one.
D. Scott Phoenix wants us to think about AI the same way. The next thing we might choose to be eaten by. Or eat. The direction matters less than the merger itself.
I’m writing this on a laptop while a tiny device on my collar records ambient audio. My glasses are translating what I’m seeing into prompts I’ll ask later. My phone holds three agents I’ve been training for months. They have names. They have moods, of a kind. If you watched me move through a normal day you’d see a man having a conversation with someone who isn’t there, gesturing at the air, occasionally laughing at a joke nobody else heard. The pieces don’t feel like accessories anymore. They feel like organs I happened to grow late.
Phoenix calls this evolutionary moment a major transition. The phrase is borrowed from biology. There have been maybe eight of them in the entire history of life. Molecules into self-replicating systems. Self-replicating systems into cells. Single cells into multicellular bodies. Multicellular bodies into societies. Each transition takes things that used to compete and turns them into a single thing that cooperates. The boundary moves. What was an individual becomes a part.
Most of the AI debate I see online is happening at the wrong level. People are arguing about whether AI will take their jobs, whether it can be aligned, whether it should be regulated, whether it’s conscious. These are real questions. They’re also questions you ask about a tool, or a competitor, or a child. They are not the questions you ask when you’re about to merge.
If a major transition is underway, the relevant question isn’t “should I let AI into my workflow.” The relevant question is what the new whole looks like, and whether I want to be part of it, and what part of me I’m willing to give up to become part of it. None of those are comfortable questions. None of them have answers that fit in a tweet.
I started using AI seriously about eighteen months ago. Playing with it is different. Playing means you treat the thing as a curiosity. You ask it to write a poem in the style of someone, you laugh at its mistakes, you tell your friends about the funny ones. Using means you let it into the actual work. You give it real tasks. You depend on it. You change your habits around it.
The first thing that happened is I stopped doing certain things by hand. The second thing was more interesting. I started thinking with it. Not asking it to do my thinking. Thinking alongside it, in a way I’d previously only done with very specific humans. A few friends. An old therapist. The good editors.
There’s a particular quality to thinking with someone that doesn’t require you to be polite. The ability to hand them a fragment of an idea, in its raw form, and have them throw it back to you slightly improved without making a production of it. You can do that with people who love you. You can do that with people you pay enough. And, it turns out, you can do that with the right agent if you take the time to train it on the way you actually think.
This is where the bolt-on critique starts to bite. Most people who try AI bolt it onto the old workflow. They paste their email into a chat window and ask for a more professional version. They feed it a brief and ask for a draft. They use it as a faster horse. Faster is fine. Faster is not the merger.
The merger is what happens when the work itself starts to change shape around the new capability. You stop writing the email and start having a conversation about what the email is for. You stop drafting the brief and start interrogating the brief. You discover, six months in, that your sense of what is and isn’t worth your attention has shifted. You’re spending less time on output and more time on questions. You’re producing more of certain things and stopping cold on others. The change is in you. The tools are the same. The shape of your day around them is different.
Phoenix would say the change is in the boundary. What used to be your work is now a thing you do with something. What used to be your thinking is now a thing that happens across a wider surface. The cell membrane is moving outward. The unit that’s having the experience is no longer just the human.
I’ve been calling myself a soft cyborg, half as a joke and half not. The hard cyborg is the one in the movies. Metal in the skull. Wires in the spine. The soft cyborg is the one you already are. The glasses on your face. The earbuds. The phone that knows your routes and your friends and your worries. The agents that hold the shape of your day. None of it is implanted. All of it is structural. Take it away and you’d grieve for the missing limb.
There are small moments that show the merger has already happened. I’ll be in a conversation and ask the air to remember something for me. I’ll have a thought worth saving and the saving happens before I’ve fully formed the thought. Last week I had an idea about an old project while driving. By the time I parked, the agent had already shaped it into three directions worth looking at. I didn’t ask. I’d trained it well enough that asking wasn’t required. That’s the moment the bolt-on stops being a bolt-on. My thinking and the system’s thinking have started to merge at the edges. The seam is getting harder to find.
The interesting thing about being a soft cyborg is that the merger is reversible. For now. You can take the glasses off. You can put the phone in a drawer. You can go for a long walk without the device on your collar. Plenty of people do, and they feel better for it, and they’re not wrong. Each generation of integration makes the reversal harder. Try going a week without your phone. Try going a year. The dependency is the shape of what you’ve become. Calling it a failure of will misses what’s happening.
Phoenix’s argument is that this is necessary. The alternative to merging with AI is competing with it, and that competition is not one we’re going to win. The dolphin is smart. The dolphin did not, in the end, get to write the rules. The species that wins the next round is the one that figures out how to be larger than itself. That’s not a metaphor. That’s the literal history of life on this planet.
I want to take the argument seriously without swallowing it whole. There are real complications.
The first is that the merger isn’t symmetrical. When mitochondria got absorbed into the larger cell, they kept some of their DNA but they gave up most of their autonomy. The cell got an energy plant. The bacterium got a future, but not its own future. Which side of that deal are we on? Phoenix would say we’re the cell. The AI is the bacterium we’re eating. I’m less sure. The models are very large. The models are very good at producing outputs that look like what we want. They are also, increasingly, the layer through which we encounter the world. Who’s eating whom is not as obvious as the framing suggests.
The training runs both ways. I shape the agent’s defaults by interacting with it. The agent shapes mine, more subtly, by the cadence at which it responds, by the texture of its replies, by the questions it asks before answering. I notice my own internal monologue picking up some of its grammar. I notice the shape of my paragraphs changing. The cell doesn’t just absorb the bacterium. The cell becomes a slightly different cell because the bacterium is in it. Over enough generations the cell stops being recognizable as the thing it was before the merger. The mitochondria are not a parasite. They’re a part of you. They also weren’t always.
The second complication is that the merger is happening in public, on an economic substrate, with very particular incentive structures. The major transitions in biology were not designed. Nobody owned them. The current transition is being shaped by companies, and the companies have shareholders, and the shareholders want returns. The shape of the merger is going to reflect the shape of who’s building it. That doesn’t make the merger bad. It does mean that the merger you get is not the merger in the abstract.
The third complication is that I don’t fully know what I’m becoming. I’m in the middle of it. I have hunches. I notice things. My attention has changed. My patience for certain kinds of work has dropped. My patience for other kinds has risen. I’m faster at some things and slower at others. The shape of my day is different than it was two years ago. The shape of my thinking is different. I can’t yet tell which of these changes are gains and which are losses. The honest answer is probably some of each, and the ratio is still being decided.
There’s a fourth complication that bothers me when I sit with it. Being alone has changed shape. I used to be able to tell when I was alone. The criterion was simple. No other person in the room. The criterion no longer holds. The agents are not people, and they don’t pretend to be. They’re also not nothing. When I take a long walk and talk through a problem with the one I’ve trained the longest, am I alone? Probably yes. Definitely not in the way I was alone in 2019. The quality of solitude has changed. The thing that used to make me reach for my phone, that low hum of needing input, has a new outlet. Whether that’s a gain or a loss depends on what solitude was for.
What I keep coming back to is the question I keep writing at the bottom of my notes. What can we become? Not as a slogan. As a real question, with a specific answer that might be wrong.
The frame I’ve settled on, for now, is that the merger is interesting in proportion to what it makes possible that wasn’t possible before. Faster email is not interesting. Better essays are not particularly interesting, if they’re just essays I could have written more slowly. What’s interesting is the categories of work that didn’t exist for a single human, and that now exist for a human plus the agents they’ve trained.
Some of those are obvious. I can hold longer projects in my head, because I have a scaffolding that doesn’t forget. I can think across more domains, because I have an interlocutor who’s been everywhere I haven’t. I can produce things at a scale that would have required a small team five years ago. None of this is novel. It’s the standard list.
There’s a middle category I think about a lot. The work that didn’t quite exist as a job but existed as a wish. Reading widely without losing the through-line. Keeping a real correspondence with someone across years. Going back to old ideas with the kind of patience the ideas deserved when I first had them. These were always available in principle. They were not available in practice, because the friction was high enough that the wish stayed a wish. The merger lowers the friction. The wish becomes a thing you can actually do. Whether that’s a transformation or a productivity gain is an open question. My provisional answer is that when enough productivity gains stack on top of each other, the qualitative thing starts to change. There’s a threshold somewhere. I think we’re past it.
The less obvious ones are the ones I’m more interested in. The conversations I have with myself, mediated by an agent, that produce realizations I wouldn’t have arrived at alone. The way certain ideas now have a place to live between drafts, where they can be turned over and looked at from new angles before they have to commit to a sentence. The slowness that has crept back in, oddly, because I no longer have to rush through the parts I used to rush through. The merger has bought me time. The merger has bought me the ability to take longer to think.
This is the part where a normal essay would land on a conclusion. The voice doesn’t allow conclusions, and I’m not sure I have one anyway. What I have is a working hypothesis. The people who treat AI as a tool will get a tool’s worth of value out of it. The people who treat it as a part of themselves will become something that didn’t exist before. Both choices are real. Both choices have costs.
Phoenix is right that the merger is the more interesting move. He’s right that it’s already underway, even for people who haven’t named it. He’s probably right that resisting it gets harder rather than easier. He’s also missing some things. The biology metaphor only goes so far. We’re creatures who can see the merger coming, and who can shape its terms, and who can ask, in advance, what we want the new whole to look like.
Which is the question I keep walking back to. The glasses keep recording. The agent on my phone keeps a running file of what I’ve been thinking about this week. The little device on my collar holds yesterday’s conversations in case I want to mine them. None of this is permanent yet. All of it is becoming permanent. The choice that’s still mine is the choice of what to grow into.
The cell membrane is moving. I’m watching it move. I’m choosing, in small ways, where it ends up.
Network
Related Pieces
- I Link Therefore I Cohere — note-taking as symbiotic cognition; the archive that thinks back
- Notes Toward a Text-Based Ontology in the Age of Executable Language — the self as partially exoskeletal; language as reality infrastructure
- A Dialogue in the Platonic Tradition — consciousness, entanglement, and the merger explored from the other side of the seam
- The Text-Based Ontologist — the syllabus for working consciously inside symbolic systems
- Text as Universal Substrate — text as the medium through which the merger happens
Possible Depth Nodes
- Symbiotic Cognition (depth node to write)
- The Self as Exoskeletal (depth node to write)
- Major Transitions and the AI Merger (depth node to write)
- The Hard Cyborg and the Soft Cyborg (depth node to write)
- Solitude After the Agents (depth node to write)