Atoms, not algorithms.

IThe Prelude

For three years the boardroom conversation has been about models: which one to license, how many agents to deploy, what the inference bill will be next quarter. That attention is deserved. It is also incomplete.

Software reshapes how decisions get made, but it rarely changes what a factory can make. The industrial revolutions that rewrote economies did exactly that. Steam changed what a mill could spin. Electricity changed what a line could stamp. Petrochemistry changed what a plant could mold. My argument is that the next revolution will be built from molecules, not models, and that artificial intelligence will matter most as the instrument that lets us design them.

IICells as Factories

Biology is becoming an engineering discipline because reading and writing its code has become cheap. The Human Genome Project took thirteen years and billions of dollars. By 2024 high-throughput platforms could sequence a human genome for about $200. That collapse, faster than Moore’s law, turns the living cell into a programmable unit of production.

Industry is already using it. In 2019 Merck and Codexis replaced a long chemical route to the HIV drug islatravir with a cascade of nine enzymes, five of them engineered, running in a single aqueous stream. There were no purifications between steps and only one isolation, of the drug itself, at a 51 percent overall yield. The process looked less like a chemistry bench than a fermentation tank following a recipe.

The ceiling is high. The McKinsey Global Institute estimated that as much as 60 percent of the physical inputs to the global economy could, in principle, be produced biologically, from fuels and plastics to fibers and food. The same analysis cautions that reaching that potential is a long way off. Even modest progress, though, would redraw supply curves across much of what industry buys.

IIIMatter by Design

The second front is materials, and here the models finally earn their keep. Nanotechnology gives us the ability to place and pattern matter at the scale of atoms, in catalysts, membranes, coatings, and battery electrodes. What it has lacked is a fast way to know which arrangements are worth making.

That is changing. In 2023 Google DeepMind’s GNoME system predicted 2.2 million new inorganic crystals, about 380,000 of them stable, lifting the count of known stable materials from roughly 48,000 to 421,000. In a parallel study, a robotic laboratory at Lawrence Berkeley National Laboratory synthesized more than forty of the predicted materials with minimal human hands.

Prediction is not production. Few of those crystals have been made, and fewer still at volume. That gap is the point. The bottleneck has moved from discovering what is possible to manufacturing it reliably, at scale, at a cost a customer will pay. That is a problem builders know how to solve.

The bottleneck has moved from discovering what is possible to manufacturing it reliably, at scale, at a cost a customer will pay.

IVWhy Builders Should Pay Attention

If value migrates from code to chemistry, it migrates toward the people who own reactors, fermenters, clean rooms, and supply chains. An algorithm can be copied at almost no marginal cost. A qualified bioprocess, a validated plant, and a trained operating crew cannot. Physical capability is becoming defensible in a way that software increasingly is not.

Three implications follow for industrial leaders.

First, the feedstock map will change. A plant that buys petroleum-derived inputs today may soon compete against one that grows them from sugar, or one that substitutes an engineered material for a heavier conventional one. Procurement strategy becomes technology strategy.

Second, the scarce skill will be scale-up. Turning a promising strain or crystal into a consistent product is unglamorous work in process engineering, quality systems, and regulatory affairs. It is also where most ventures in this space stall, and where experienced manufacturers hold an advantage they routinely undervalue.

Third, digital investment should be judged by what it lets a company make. The most valuable AI platform in a manufacturing business may not be the assistant on the intranet. It may be the model that shortens a formulation or materials development cycle from years to months.

VThe Work Ahead

None of this will arrive on a press-release schedule. Biology is heavily regulated, rightly so, and self-replicating systems carry risks that software does not. Public trust will set the pace as much as science will.

The direction, however, is clear. The companies that lead the next era will treat models as instruments and molecules as products. The engineers, operators, and plant managers who have spent careers making physical things work at scale are not being left behind by this revolution. They are the people it will need most.

1. Genome sequencing review, Labiotech
2. Design of an in vitro biocatalytic cascade for the manufacture of islatravir, summarized by BioCentury
3. In-vitro multi-enzymatic cascade reactions, University of Illinois
4. The Bio Revolution, McKinsey Global Institute
5. Google DeepMind unveils GNoME, Maginative
6. Google AI discovers 2.2 million new materials, Freethink