In the past few years, with the rise of in vivo cell and gene therapy, the industry has become accustomed to comparing it to mature in vitro therapies. The logic of this way of thinking is that since in vitro CAR-T therapy has achieved success through strict production process control, in vivo therapy simply replaces the delivery carrier, and the development path should be similar.

But this analogy is showing its limitations.

The core of in vitro therapy is that cells undergo screening, amplification, and quality inspection in vitro, and uncertainty is eliminated as much as possible before entering the patient. This method is indeed effective in relatively controllable low entropy environments such as blood tumors. However, when the therapy enters the body, the cells no longer face a culture dish, but a life system full of heterogeneity, feedback loops, and state transitions.

When ReHeart's myocardial patch is attached to the surface of a failing heart, and Amchepry's dopamine neurons are injected into the trembling brain, these cells perceive the local microenvironment, respond to physiological signals, and interact complexly with host tissues. They are no longer objects that are "manipulated", but "operators" who actively participate in regulation.

The mechanism of action of ReHeart precisely illustrates this point: it does not directly perform contractile function, but works by sensing the ischemic environment, secreting repair signals, and mobilizing the host's own regenerative potential.

The problems faced by Amchepry are even more complex. How do implanted dopamine neurons release dopamine at the correct time, location, and intensity? This is no longer an execution issue, but a decision-making issue.

The environment faced by the in vivo treatment system has several fundamental characteristics: spatial heterogeneity (different tumor microenvironments and tissue gradients); Time variability (disease progression, sustained changes in immune activation); Cell state dependence (depletion, constantly changing activation thresholds); The irreversibility of many processes (once entering certain trajectories, the cost of reversal is extremely high).

In such an environment, the core issue is no longer whether intervention can be accurately executed, but whether intervention should occur, when it should occur, and under what conditions it should occur. This has shifted from an execution problem to a decision-making architecture problem.

Observing the clinical outcomes of CAR-T and gene therapy in recent years, a recurring pattern can be observed: initial biological activity appears, partial efficacy is achieved, followed by unexpected toxicity, persistent loss, or disease recurrence, and ultimately the system stabilizes in a suboptimal state.

These results are often attributed to insufficient carrier efficacy, weak efficacy, and improper target selection. But in many cases, these explanations describe 'what' went wrong but fail to clarify 'why the system went wrong'.

In an in vitro system, protective mechanisms are embedded, and production checkpoints, release standards, and process control collectively constitute constraints. In the internal system, these constraints basically disappear. When the activation, expansion, and differentiation of cells occur without sufficient perception and permission, therapy may push the biological system into a seemingly bright track in the early stages but difficult to exit in the later stages.

Excessive immune activation leads to toxicity, while attacking tumors leads to self exhaustion. These are not issues of execution quality, but rather logical decision-making about when to take action.

In the face of failure, common coping strategies include increasing abilities: higher doses, more potent carriers, and additional genetic modifications. These methods may improve short-term efficacy, but they also often amplify long-term instability.

If there is a lack of mechanisms to perceive biological states, a lack of permission logic for intervention, and a lack of measures to stabilize outcomes, more powerful tools will only drive the system to run faster and sometimes fall into more complex dilemmas. In high entropy environments, if abilities are not constrained, they can actually increase systemic risk.

There is often a debate in the industry about "virus or non virus, in vitro or in vivo, single or multiple administration". But these debates may have bypassed the real problem. The core is not about which modality is superior, but whether the treatment system is designed to manage the biological state transitions that change over time.

This requires a shift from "tool thinking" to "architecture thinking", with perception, decision-making, collaboration, and stability as the core elements of design. When therapy is designed as a managed system rather than isolated intervention, the focus of the design shifts.

State perception becomes a prerequisite, as cells not only need to recognize targets, but also need to perceive whether the biological environment is ready and the surrounding conditions; Actions require permission rather than automation, and intervention should not be triggered by a single static input, but should occur when conditions permit; Timing is a variable that needs to be managed, and the timing of an action may be equally important as the action itself; Stability is a clear goal, not only inducing change, but also guiding the system towards a sustainable and healthy state.

Japan's' conditional and time limited approval system 'acknowledges this complexity from a regulatory perspective. The seven-year data collection period means that it is impossible to fully predict the long-term behavior of cells in vivo in vitro, and the uncertainty has not disappeared, but has shifted from the production end to the clinical follow-up end.

Looking back at the two products approved in Japan, their design logic already includes certain elements of "governance thinking".

ReHeart did not choose to directly replace myocardial cells, but instead improved the microenvironment by promoting angiogenesis, which means that cell behavior is regulated by the local environment rather than being completely programmed.

Amchepry's dopamine neurons need to integrate with the host neural circuit and receive control from upstream neurons, essentially constructing a decision-making framework that constrains cellular output based on physiological needs.

But to truly achieve the design at the "governance" level, further progress is needed, and state perception should become a standard capability of cell therapy. Future iPS cells not only need to express therapeutic molecules, but also need to have the ability to sense inflammation levels, hypoxia levels, and neural activity patterns.

The proliferation, differentiation, and secretion of cells should be permitted by environmental signals rather than automatic execution of preset programs. In the treatment of Parkinson's disease, the timing of dopamine release may determine the degree of improvement in motor function more than the total amount released. Implanted cells not only need to function, but also need to reach a sustainable coexistence agreement with the host system to avoid falling into a trajectory of exhaustion or abnormal transformation.

For the development team, these thoughts can be translated into specific operational adjustments: preclinical studies not only need to answer whether the construct is effective, but also need to answer under what conditions it should not be activated. The translational medicine stage not only requires assessing the strength of early signals, but also determining whether they point towards a sustainable trajectory. Clinical interpretation needs to distinguish between "early success" and "trajectory alignment". During project review, some failures are due to execution issues, while others are due to architecture problems.