A Good Enough AI Is Already Transforming the Economy

The Analysis’ Foundation and Argument

“When Will AI Transform the Economy?,” asks Andre Infante. His analysis at substack offers an explanation for the slow economic transformation despite fast AI advancements. The central question of his investigation is: Why hasn’t AI changed the economy yet, and when will it? Infante identifies the core problem in the discrepancy between the impressive peak performances (”intelligence ceilings”) and the lack of everyday reliability (”intelligence floors”) of current models.

The articles conceptual model is based on three fundamental, implicit assumptions. First, the economic value of AI is primarily understood as a function of its technical reliability (A1). The central question is, “how often does the model fuck up?”. Second, it is assumed that the transformation proceeds in clearly definable, predictable phases (A2): a long “hybrid phase” of gradual efficiency gains is followed by an abrupt jump into a “post-labor regime.” Third, the analysis posits that the future progress of AI capabilities can be reliably predicted by extrapolating current benchmark trends (A3). The analysis relies on trend lines from organizations like METR, viewing them as the “best signal we have” for the coming years. These three assumptions form the foundation for the analysis’s conclusions.

Infante’s Central Results and Findings

The central finding is that the current generation of AI is more “impressive than it is useful” due to its error-proneness. This lack of reliability (A1) makes it unsuitable for mission-critical business processes and explains the missing productivity revolution.

Consequently, Infante argues, the real transformation will only be triggered by achieving near-perfect accuracy. Only when AI is so reliable that human oversight becomes obsolete can the abrupt transition to the “post-labor regime” occur, in which entire professional groups are replaced. This is the direct consequence of the assumption of discrete phases dependent on accuracy (A2). Based on the extrapolation of benchmark data (A3), this turning point is located in the early to mid-2030s for many areas of knowledge work. The implication is a form of strategic waiting: the great upheaval will only come when the technology reaches a quasi-perfect level of maturity.

A Systems-Theoretical View and Alternative Conclusions

The observations presented by Infante are consistent within his assumptions. However, a view from a systems-theoretical and nanoeconomic perspective shows that these underlying assumptions steer the conclusions in a misleading direction. His interpretation will be critically examined below.

The first assumption—the focus on technical reliability—is a category error. From a nanoeconomic perspective, the value of an AI is not its abstract accuracy, but its ability to deliver a marginal but decisive information improvement (ΔInfo) at its point of use within a company’s workflow, a Critical Information Point (CRIPO) (Gerlach, 2025). Every action in a workflow follows a clear logic: Information → Decision → Action (the IDEA mechanism). A seemingly “dumb” but perfectly placed and integrated AI that can deliver decision-relevant information of sufficient quality at a business bottleneck today can create more value than a highly developed AI whose accurate output is only available in the distant future. Value is not created by technology itself, but by its careful integration and use within a value-creation process. This is reminiscent of Herbert Simon’s satisficing principle.

For human users, AI is already a “good enough” tool in everyday contexts, as these examples illustrate:

• An email draft generated by an AI is probably not literarily perfect, but it is good enough to save 80% of the writing effort if the correction and prompting can be done in one-fifth of the previous time.
• An AI-generated summary of a long document may not capture every nuance, but it is good enough to convey the key points in a fraction of the reading time. Potentially relevant sections can be explored in more detail as needed. The total effort is incomparable to the normal time required.

The second assumption—discrete, linear phases—contradicts the nature of complex adaptive systems. The economy is not a mechanism that flips from one state to another. It is a network in which local changes trigger cascading feedback loops, a phenomenon we call the AI-Butterfly-Wheel-Effect (Gerlach, 2025). This leads to a continuous, uneven reorganization of production processes (nanoeconomically: the AI-Fusion-Effect) instead of an abrupt “post-labor” jump. The sharp edges in Infante’s model are artifacts of simplification.

The third assumption—predictability through benchmarks—ignores the systemic hurdles of the real world. As W. Ross Ashby formulated, “only variety can destroy variety.” The complexity of a real business environment exceeds that of standardized tests by orders of magnitude. The bottleneck for transformation is not raw AI performance, but the extremely slow and costly co-evolution of processes, data infrastructure, organizational structures, and human skills. The ability of humans to adapt their workflows—in other words, AI integration—is the bottleneck.

The Economic Transformation Induced by “Good Enough” AI

If one considers Infante’s data under these more robust assumptions, a different picture emerges.

The low speed of economic transformation is not the result of insufficient AI reliability, but an emergent property of a system that responds to local, informational leverage rather than global technology upgrades.

From this, it follows, first, that waiting for “perfect AI” is a losing strategy. In the meantime, first movers will have used AI’s potential to identify and eliminate countless inefficiencies in their workflows. This satisficing approach leads to gradual improvements that can ultimately result in significant leaps in quality. The integration and use of AI in workflows are crucial for the success of local transformations of companies and markets, and thus for the changes in the economy as a whole.

Second, it means that the most important entrepreneurial skill of the future is the permanent, systematic, and pragmatic identification of those critical decision points where improved information offers the greatest leverage for workflow improvement and, consequently, business success.

Once the integration of good enough or satisficing AI into business workflows has progressed, emergent effects will become apparent in the economy. Supply chains will adapt, and jobs will change. Above all, however, better information should ultimately improve product quality and thus trigger new effects. Every small product improvement reduces the opportunities for other products to fix its weaknesses. Let’s give AI integration a little more time.

Sources:

Ashby, W. R. (1956). An introduction to cybernetics. Chapman & Hall.

Gerlach, Silvio, The Simple Nanoeconomics of AI: An Economic World Model for Exploring AI Impacts (August 13, 2025). Available at SSRN: http://dx.doi.org/10.2139/ssrn.5394649

Infante, A. (2025, October 21). When will AI transform the economy? Whatever Remains. Retrieved October 23, 2025, from https://substack.com/@andreinfante/p-176653884

Simon, H. A. (1996). The sciences of the artificial (3rd ed.). The MIT Press.