Your brain is not one thing. It never was. And if you have spent any time reading about AI systems that supposedly mimic “the brain,” that single sentence should make you pause.
The mainstream story about brain development goes something like this: a sheet of early cells forms, then gradually specializes, folding and dividing into the regions we name later — forebrain, midbrain, hindbrain. One origin, branching outward. Tidy. Intuitive. The kind of story that fits neatly into a diagram.
A 2026 paper in Nature Neuroscience by Rayyan T. Jokhai, Carolyn E. Dundes, H.S. Ahsan and colleagues points somewhere else. Their finding, titled “Two parallel neural ectoderm progenitors contribute to the developing brain,” describes two separate populations of neural ectoderm progenitors running in parallel. One gives rise to the forebrain and midbrain. The other gives rise to the hindbrain. Not one source splitting in two. Two sources from the start.
Why a Person Who Explains AI Agents Cares About Embryos
I write about AI agents for people who do not write code. So why am I reading developmental neuroscience papers?
Because the word “brain” does an enormous amount of quiet work in how we talk about AI. Every time someone describes a model as brain-inspired, or an agent architecture as neural, they are borrowing authority from biology. And the version of biology they are borrowing from is usually the tidy one — the single origin, the unified system, the one big network that learns everything.
The 2026 work suggests that the tidy version is not how the thing actually got built.
If two lineages contribute in parallel to different brain regions, then the organ we point to when we say “intelligence lives here” is closer to a merger than a monolith. Related earlier work from Dundes and colleagues in 2025 tested exactly this hypothesis — that early neural ectoderm cells are already committed to producing either forebrain/midbrain or hindbrain — using complementary approaches in living embryos. The commitment appears to happen earlier than the tidy story allows.
What This Changes About the Metaphor
Here is where it gets genuinely interesting for anyone trying to understand AI agents.
The dominant popular image of an AI system is a single large model that does everything. One set of weights. One training run. One mind. People reach for brain metaphors to describe it because a brain feels like one unified organ.
But the actual architecture that most working AI agent systems use looks nothing like that. It looks like this:
- Separate components with separate origins, built at different times
- A planning layer that decides what to do next
- A retrieval layer that fetches relevant information
- Tool-calling machinery that talks to outside systems
- A coordination layer that keeps the parts from talking over each other
Practitioners sometimes apologize for this. The assembled, multi-part nature of agent systems gets treated as a compromise — what we build because we cannot yet build the one true unified model.
The developmental biology finding makes that apology look unnecessary. Parallel construction with distinct lineages producing distinct regions is not a workaround. In brain development, according to this research, it is the actual method.
Careful With the Analogy
I want to be honest about limits here, because overreaching is how bad tech writing happens.
This paper is about embryonic cell lineages in brain development. It is not about AI. The authors are not making claims about machine learning architectures, and I am not going to put words in their mouths. Developmental origin and functional organization are different questions, and a finding about the first does not automatically settle the second.
What I am saying is narrower. The metaphor many of us use to think about intelligent systems — one unified organ, one origin, one smooth whole — turns out to rest on a picture of brain development that this 2026 research complicates. The paper was received in November 2025 and published in September 2026.
The Practical Takeaway
If you are trying to understand AI agents without a technical background, you have probably been told that the goal is one model that does everything, and that today’s patchwork of tools and components is a temporary stage.
Be a little skeptical of that framing. Systems built from parallel parts with different origins are not obviously inferior to systems built from a single source. Biology’s most celebrated example of intelligence appears, by this account, to have been assembled that way.
So the next time someone tells you an AI system works “like a brain,” you are entitled to a follow-up question. Which brain? Built how? By how many crews?
Those questions are more useful than they sound. They push past the metaphor and toward the architecture, which is where the interesting decisions actually get made.
🕒 Published: