I named my project in about thirty seconds, and it took an academic literature review to show me what that thirty seconds cost.
Early in building my knowledge-driven development framework, I needed a name. I asked Claude. It gave me around ten suggestions. One of them, Lattice, stopped me cold, because it described exactly what I was building: a structure of connected nodes, every piece linked to the pieces it depends on. The word fit like it had been waiting for the idea. I picked it and went back to work. I didn’t search for it. I didn’t check if anything else carried the name. It felt right, and feeling right felt like enough.
It wasn’t.
Weeks later I ran a proper review of the problem I was solving, reading the prior work on requirements traceability so I could state honestly what was new. That’s when I found a project called lat.md. Same problem space, different approach, and here’s the part that made me sit up: when its author described the tool, the word he reached for was “lattice.” Agent Lattice, a knowledge graph for your codebase. Different design, overlapping goal, a name that was a sibling of mine.
Coincidence, I told myself. Then a few days later I found another project. Same goal. Same name. Lattice, again. And once I looked harder, there were more “lattice” knowledge-graph tools out there than I want to admit. The word wasn’t mine. It was in the water.
So, why did three people, working separately with no contact, land on the same name for the same idea at the same time?
The answer is the actual point of this post. All three projects were built with Claude as a coauthor. We didn’t converge by accident. We asked the same model the same kind of question, and a model is not a random name generator. It’s a compression of how language tends to go. Ask for a name for “a structure of connected, interdependent nodes that hold knowledge,” and “lattice” sits near the center of the probability mass. It’s a genuinely good answer. That’s the problem. It’s such a good answer that the model hands it to everyone who asks, and none of us can see the others in the room.
This is new, and worth naming clearly. When a generation of builders ideates with the same handful of models, those models become a hidden convergence engine. Not a thief of originality, something stranger: a quiet force pulling independent people toward the same words, the same framings, sometimes the same ideas, with nobody copying anyone. You feel like you made a choice. You and a hundred strangers made the same choice, for the same reason, from the same source.
The naming was the visible symptom. The deeper version scares me more. If the model nudges three of us to the same name, it’s nudging us toward similar architectures and similar problem framings too. The convergence doesn’t stop at the label. The label is just where it’s easy to spot.
There’s a reflex that says the lesson is “don’t let AI name your thing.” Wrong lesson, and the same mistake I once made in reverse. Early in my career I almost rebuilt something that had existed for decades, because I didn’t search before I built. The fix was never to stop using tools. It was to check the tool’s output against the world before betting on it. A model handing you “lattice” is no different from a senior engineer handing you “lattice.” A strong suggestion from a source that has seen a lot. Not a verification that the name, or the idea, is yours to take.
So I built the check into my process, as a reusable skill instead of a resolution I’d forget by the next project. It does three things. It suggests names for an idea. It validates each candidate against existing projects, products, and packages, instead of trusting that a good-sounding name is a free one. And it widens past the name to the idea itself, looking for prior work with the same goal, so “has anyone built this already” gets asked at the start, when the answer can still change what you build.
That last part matters most. The name collision was a gift. A cheap, visible signal of an expensive, invisible truth: I hadn’t actually checked whether my idea was new. The duplicate name dragged me into the literature, and the literature is where I found what was genuinely mine to contribute. If the names had been different, I might have gone months longer believing I was alone in a space that was already crowded.
The tools are getting better at giving us the right answer. That’s exactly why the discipline has to shift. When the answer is cheap and good and identical for everyone who asks, the value isn’t in getting it anymore. It’s in checking it against a world the model can’t see from inside the chat.
I still like the name Lattice. I just no longer think I chose it.
Open question I’m sitting with. If the model converges our names and our framings, what’s the move? Reject its first suggestion on purpose and take the less-crowded second-best? Or accept that we’re all building in a shared current now, and compete on the execution the model can’t do for any of us? I don’t have the answer yet.