For most of human history, biology was something we could observe but not directly control. We could study how organisms developed, identify the causes of certain diseases and develop treatments that altered biological processes. Yet the systems themselves remained extraordinarily difficult to predict. Changing a single gene could have consequences across multiple tissues, and the same genetic variation could produce different outcomes in different individuals.

This is beginning to change. Advances in genome editing, synthetic biology and computational modelling are giving us increasingly precise ways to modify living systems. We can edit DNA, design genetic circuits and engineer cells to perform specific functions. The ambition is no longer simply to understand how biology works, but to develop the ability to design what it does.

The comparison with computer engineering is tempting. In software, a programmer can modify a function and test how the change affects a program. Biology appears to offer something similar. Genes encode proteins, regulatory sequences influence gene expression and molecular networks determine how cells respond to their environment. If these relationships could be understood well enough, biological systems might eventually be engineered with the same predictability as other complex technologies.

But living systems are not conventional machines. A computer generally executes instructions within a defined architecture. A cell continuously responds to its environment, regulates its own activity and adapts to changes in its internal state. Genes interact with one another through networks that vary between cell types and developmental stages. A modification that produces the desired effect in one context may have unexpected consequences in another.

This is one of the central challenges of genetic engineering. The ability to edit a sequence does not automatically provide the ability to predict its effects. CRISPR can make targeted changes to DNA, but the resulting phenotype depends on much more than the edited sequence. Regulatory networks, epigenetic states, cellular metabolism and environmental conditions all influence the outcome.

The problem becomes even more difficult when the objective is to engineer a complex trait. Many characteristics relevant to human health are influenced by thousands of genetic variants and their interactions with environmental factors. Ageing, metabolic disease and cognitive function cannot be reduced to a single gene that can simply be switched on or off. They emerge from interacting biological processes operating across different scales.

Artificial intelligence may help bridge this gap. Models trained on genomic sequences, molecular structures and gene expression data can identify patterns that are difficult to extract using conventional methods. Some systems can propose new protein sequences or predict the effects of genetic perturbations. Others can help identify regulatory elements and prioritise experiments.

However, prediction is not the same as control. A model may identify a sequence associated with a particular function without accurately predicting what that sequence will do inside a living cell. Even a successful laboratory result may not translate into a predictable outcome in an organism. The real test is whether computational predictions can repeatedly produce the intended biological effect under relevant conditions.

This suggests that the future of programmable biology will depend on more than increasingly powerful models. It will require a continuous relationship between computation and experimentation. Models will propose modifications, experiments will test their consequences and the resulting data will improve subsequent predictions. Over time, this process could make biological engineering more systematic and less dependent on trial and error.

The same principle could transform medicine. Instead of treating every patient as an isolated clinical case, we might eventually develop more detailed models of how individual biological systems respond to different interventions. Genomic information could help identify inherited predispositions, while transcriptomic and epigenetic measurements could reveal aspects of a person's current biological state. Longitudinal observations could show how these systems change over time.

The goal would not necessarily be to create a complete simulation of a human being. Such a model may be impractical for the foreseeable future. A more realistic objective is to build models that are sufficiently accurate for specific decisions. A system that reliably predicts whether a particular intervention will improve a defined biological process could be clinically valuable without explaining every aspect of human physiology.

There are also important limits to what should be engineered. Increasing our ability to modify living systems raises questions about safety, ecological consequences and the distinction between treating disease and enhancing human characteristics. Some interventions may have effects that are difficult to reverse. Others could create inequalities if their benefits are available only to a small proportion of the population. Technical capability alone cannot determine which applications are desirable.

Programmable biology should therefore be understood as a gradual shift rather than a single technological breakthrough. The first systems will be limited to specific proteins, cell types and biological processes. Their success will depend on how reliably they behave outside carefully controlled experiments. Progress will come from discovering which parts of biology can be engineered predictably and which remain too complex for current methods.

The long term ambition is to move from describing biological systems to developing reliable ways of changing them. Achieving this will require a deeper understanding of the relationship between genetic information, cellular state and physical function. It will also require better experimental infrastructure and methods for validating predictions.

Biology may never become programmable in the same sense as software. Living systems possess a complexity and adaptability that resist complete control. But if we can learn to predict their behaviour within defined limits, we may gain something more useful than the ability to edit DNA. We may develop the ability to design biological interventions with measurable, reproducible and increasingly predictable outcomes.