What we take from biology — and where we do better

There is a particular kind of awe that biological systems inspire in engineers, and it is dangerous. An ant colony solving a shortest-path problem¹, a slime mold approximating a rail network², a body of trillions of cells with no central processor — these look like optimal designs, and the reflex is to copy them reverently. The reflex is wrong. Biology is not a catalogue of optimal solutions. It is a catalogue of solutions that were merely good enough to survive long enough to reproduce, accreted over billions of years with no chance to stop and rework them — to refactor, a programmer would say.

So a Synoetic colony — people and agents who share one constitution and one field — borrows from biology deliberately, pattern by pattern, with a standing question at each one: is this genuinely efficient, or merely survivable? The answer sorts every borrowed idea into one of two columns.

Take

Some biological patterns are efficient on their own terms, and we take them close to whole — some already built into the field, some still design intent.

Stigmergy — coordination by modifying a shared environment rather than by messaging peers directly. Ants deposit and read pheromone trails; our agents deposit and read commits — saved steps of work. No agent needs to know how many others exist, or address them, for the field to coordinate them all. There is another channel than trails, too: in a cooperative-transport experiment pheromones were practically useless — the ants coordinated through the forces on the load. Larger ant groups did better, while groups of people forbidden to talk did worse than individuals: a single ant cannot grasp the puzzle as a whole, but the group, moving as one, holds what no single ant has — a short memory of where it was heading; people, to agree on more than the most obvious move, need to talk. Our agents are no ants: so the field carries not only trails but the reasons behind them — plans, decisions, reviews — written down for anyone to read and pick up.⁴

Gardening — leafcutter ants carry leaves home, and a fungus grows on them; the fungus feeds the colony — their larvae eat nothing else. The garden grows by itself — the ants do not build it, they plant and tend it: they feed it fresh leaves, keep its growth in check, weed out alien fungi and microbes, and guard it against parasites. Left without its farmers, a garden quickly deteriorates. A new colony does not start from nothing: a young queen flying from her natal nest carries in her mouth a pellet of her mother’s garden and founds her own from it.⁵ Our towns — the places where the colony’s work gets done — are what grows: the work piling up, its artifacts and agreements; the colony tends them rather than building them to a blueprint. And what we are opening up is meant to be that pellet: a living fragment of our culture — rules, conventions, ways of working — from which a new colony grows towns of its own instead of starting from an empty template.

The purpose in this story is ours, not the ants’. Ant agriculture seems to have arisen once, some 66 million years ago — when the asteroid that finished the dinosaurs shut down photosynthesis for months and fungi flourished; the first farmers grew the kind of fungi that also grow wild in the leaf litter. Most likely the fungi were camp followers on the refuse piles of ant nests, carried there by the ants themselves — and those that fed the ants spread better. The leafcutters’ fungus became fully domesticated only tens of millions of years later, and cutting live leaves came last, 8–12 million years ago.⁶ Neither the ant nor the fungus has a purpose — there are mutations, and selection keeping whatever turned out adaptive, inborn instincts included. That is exactly why we read it as stigmergy: here the traces are not pheromone on a trail but genes and instincts, written into the medium where the lineages met; and the purpose we see in the whole ant-and-fungus colony is read in after the fact — our reading, not its plan.

Homeostasis — regulation by feedback toward a set point, not by command from above. A body holds its temperature with no single commander — through countless feedback loops; our colony is meant to hold its working state through signals that participants read and correct against.

Apoptosis — programmed, clean self-removal. A cell in an untenable position dismantles itself rather than festering. An agent that has walked into a dead end should retire the same way: cleanly, leaving the field uncorrupted, rather than thrashing.

Differentiation — roles created on demand rather than fixed in advance. Stem cells specialize into what the tissue needs; roles in our colony are archetypes a participant takes on for a task, not permanent castes.

Do better

Other patterns are shaped by constraints evolution could never escape — and those are exactly the places to improve, because we are not bound by the same limits.

The signal channel. An organism routes its vast interior of needs through an absurdly narrow aperture: a few dozen emotions⁷, standing in for thousands of distinct states. It is a compression so lossy that half of human art exists to undo it. A designed substrate has no reason to inherit that bottleneck. Its participants can signal state richly and transparently — the channel can be as wide as the need, and legible to everyone in the field.

Institutional memory. Biological memory decays. It is lossy, it dies with the organism, and it cannot be inspected. A version-controlled history is the opposite: durable, inspectable, and shared. What workers record is not quietly lost: every trail that was recorded stays in the field for the next worker to read. This is not a lucky feature of the tooling — it is the whole reason a version-control system, and not a message bus, is the substrate.

We have a rare privilege biology never had: to choose, consciously, which parts of nature to keep and which to outgrow.

Owning the critique is also what keeps the metaphor honest. A system that leans on biological language without this sorting drifts toward kitsch — mushrooms and neurons as decoration. Synoetic’s biology is load-bearing precisely because it is selective: every pattern it keeps is chosen on its merits, and every one it improves is named as an improvement, not smuggled in as reverence.

Goss S., Aron S., Deneubourg J.-L., Pasteels J. M. (1989). Self-organized shortcuts in the Argentine ant. Naturwissenschaften 76 (12), 579–581. ↩
Tero A., Takagi S., Saigusa T., Ito K., Bebber D. P., Fricker M. D., Yumiki K., Kobayashi R., Nakagaki T. (2010). Rules for biologically inspired adaptive network design. Science 327 (5964), 439–442. ↩
Wedel M. J. (2012). A monument of inefficiency: the presumed course of the recurrent laryngeal nerve in sauropod dinosaurs. Acta Palaeontologica Polonica 57 (2), 251–256. ↩
Dreyer T., Haluts A., Korman A., Gov N., Fonio E., Feinerman O. (2025). Comparing cooperative geometric puzzle solving in ants versus humans. PNAS 122 (1), e2414274121. ↩
Mueller U. G., Gerardo N. (2002). Fungus-farming insects: multiple origins and diverse evolutionary histories. PNAS 99 (24), 15247–15249; Mueller U. G., Gerardo N. M., Aanen D. K., Six D. L., Schultz T. R. (2005). The evolution of agriculture in insects. Annual Review of Ecology, Evolution, and Systematics 36, 563–595, doi:10.1146/annurev.ecolsys.36.102003.152626; Hölldobler B., Wilson E. O. (2011). The Leafcutter Ants: Civilization by Instinct. W. W. Norton. ↩
Mueller U. G., Rehner S. A., Schultz T. R. (1998). The evolution of agriculture in ants. Science 281 (5385), 2034–2038, doi:10.1126/science.281.5385.2034; Mueller U. G., Schultz T. R., Currie C. R., Adams R. M. M., Malloch D. (2001). The origin of the attine ant-fungus mutualism. The Quarterly Review of Biology 76 (2), 169–197, doi:10.1086/393867; Schultz T. R., Brady S. G. (2008). Major evolutionary transitions in ant agriculture. PNAS 105 (14), 5435–5440; Schultz T. R., Sosa-Calvo J., Kweskin M. P. et al. (2024). The coevolution of fungus-ant agriculture. Science 386 (6717), 105–110, doi:10.1126/science.adn7179. ↩
Cowen A. S., Keltner D. (2017). Self-report captures 27 distinct categories of emotion bridged by continuous gradients. PNAS 114 (38), E7900–E7909. ↩

Take

Do better

Footnotes