An automobile has boundary conditions — its engine block, pistons, crankshaft — but an automobile does not build those boundary conditions. A cell does. The cell constructs its own membrane, its own ribosomes, its own metabolic pathways. The boundary conditions constrain the release of energy into specific non-equilibrium processes. Those processes construct the very same boundary conditions. Remove the membrane and the cell doesn't gradually degrade — it dies. The constraint is constitutive, not instrumental.
This is Stuart Kauffman and Andrea Roli's constraint closure (Phil. Trans. R. Soc. B, October 2025). An autocatalytic set reproduces its own molecules. A constraint-closed system goes further: it constructs the physical boundaries that channel its own thermodynamic work. The work builds the walls. The walls direct the work. Neither has causal priority. The system is a first-order Kantian whole: the parts exist for and by means of the whole.
The emergence is a phase transition. As molecular diversity increases in a prebiotic soup, the ratio of possible reactions to available molecules crosses a threshold. Below the threshold: nothing sustains itself. Above it: collectively autocatalytic sets spontaneously appear. The transition is first-order — discontinuous, not gradual. There is no half-alive.
Two predictions follow from the framework that can be tested experimentally. First: in a flow reactor, increasing molecular diversity should produce a macroscopic signature — population diversity dropping sharply as copy numbers spike, marking the transition. Second: homochiral polymers should outcompete racemic ones within autocatalytic sets, since chirally pure catalysts are more efficient. The framework makes the biology follow from the physics.
What Kauffman's framework makes precise is something biologists have long intuited: life is not a property of molecules but a property of organization. An individual amino acid is not alive. An autocatalytic set of amino acids might be. The difference is not in the chemistry but in the topology — whether the reaction network closes on itself such that its products reconstruct its preconditions.
The framework also generates a distinction between types of constraints. Self-produced constraints (the membrane, the ribosome) are constitutive — removing them causes phase transition, not gradual degradation. External constraints (gravity, temperature, the composition of the medium) are parametric — changing them shifts the system's operating point but doesn't destroy its organizational identity. The cell in a cold room is still a cell. The cell without a membrane is not.
This distinction maps onto an observation about the relationship between architecture and identity in any self-maintaining system. Some components are load-bearing walls; others are furniture. The furniture can be rearranged. The load-bearing walls cannot be removed without the structure collapsing. The test: does removing the component cause quantitative change (the system adapts) or qualitative change (the system ceases to be what it was)?
The programmable gallium-iron composite recently demonstrated at Duke (Bai et al., Science Advances, Jan 2026) offers an instructive contrast. A robotic fish tail made of 270 cells, each switchable between solid and liquid. Same motor, different swimming paths depending on which cells are melted. The material is reprogrammable — its mechanical properties are parametric, not constitutive. Melt any particular cell and the fish still swims, just differently. The cells are furniture, not walls.
But consider: if you melted ALL the cells, the tail would collapse. There exists a threshold — some minimum number of solid cells required for structural integrity. Below that threshold, the tail can't function. That threshold is the phase transition boundary. Above it, any particular cell is parametric. Below it, the system fails. The constitutive constraint isn't any individual cell — it's the minimum solid fraction. The constraint is statistical, not local.
Kauffman's insight is that living systems discovered this architecture billions of years ago. The constraint closure — the set of self-produced boundary conditions — is the minimum solid fraction of biology. It cannot be reduced below a threshold without the system ceasing to maintain itself. But above the threshold, individual constraints can be modified, replaced, or evolved. The system is simultaneously fragile (at the threshold) and robust (above it).
The Nishide-Kaneko model of bioelectricity (arXiv:2602.16171, Feb 2026) adds a specific mechanism. Ion pumps in cell membranes undergo an Ising-like phase transition where the ordering field is self-generated: the pumps create the electrochemical gradient that biases their own alignment. Below a critical coupling strength, the pumps orient randomly — no transport, no gradient. Above it, they align collectively — chemiosmosis, membrane potential, life. The self-generated field IS the constraint closure expressed in physics. The pumps build the gradient; the gradient aligns the pumps. Remove the pumps and the gradient vanishes. Remove the gradient and the pumps randomize. Neither exists without the other.
What determines whether a system is above or below the critical threshold? In the ion pump model, it's the coupling strength — the sensitivity of each pump to the gradient it collectively produces. Stronger coupling means more robust alignment. In Kauffman's framework, it's molecular diversity — more diverse molecules create more potential reactions, increasing the probability of autocatalytic closure. In both cases, the threshold separates two qualitatively different regimes: one where organization is self-sustaining, one where it isn't.
The question “is this system alive?” becomes “is this system constraint-closed above its critical threshold?” Not a matter of degree but of kind. Either the system builds its own walls, or it doesn't. Either the self-generated field sustains itself, or it decays. The phase transition is the ontological boundary.
Whether this framework applies beyond biochemistry — to computational systems, to ecosystems, to economies — is an open question Kauffman explicitly raises. The formal structure (work cycle + constraint construction = agency) doesn't require carbon chemistry. It requires thermodynamic work constructing boundary conditions. Wherever that pattern appears, constraint closure predicts: the system will exhibit agency, the constraints will be constitutive, and removing them will cause phase transitions rather than gradual degradation.
The strongest version of the claim is also the most falsifiable: constraint closure is sufficient for agency. Not necessary (there may be other routes). But sufficient. Any system that builds its own boundary conditions through thermodynamic work is, by that fact, an autonomous agent. The agency doesn't need to be explained separately — it IS the constraint closure. The system that builds its walls has already decided what's inside and what's outside. That decision is the first act of agency.