‘Good enough’ isn’t enough: Why a Nobel Laureate’s ‘satisficing’ model breeds human stagnation

David Epstein’s recent New York Times celebration of Herbert Simon’s Nobel Prize-winning concept of “satisficing,” or, settling for what’s good enough instead of perfect, arrives at an unusual moment.

It’s coming at a time when artificial intelligence is rewriting the rules of technological risk, when a single engineered pathogen could paralyze global civilization, and when the compounding weakness of interconnected systems is altering every domain from supply chains to financial markets. So, Epstein’s prescription is to stop searching for better answers once a “good enough” threshold is reached.

This is not wisdom. It is a civilizational sedative dressed up as cognitive hygiene.

Simon’s original insight was genuine and useful: limited rationality is real, human cognitive capacity is finite, and for low-stakes, high-frequency decisions — what to wear, what to eat for breakfast — the concept of satisficing is a sensible energy-management strategy.

The error is in the generalization. What works as a lifestyle hack for a Nobel laureate’s wardrobe becomes a structural catastrophe when embedded into the architecture of complex systems.

The core vulnerability of the satisficing model is its assumption that decisions are isolated events. In complex, interconnected architectures — whether global supply chains, financial markets, or technical infrastructure — decisions are sequential and interdependent.

When a manager accepts a 95% satisfactory threshold at step one, a 5% margin of error is introduced. Across five sequential decisions, each accepted at the same threshold, overall system integrity degrades to roughly 77%. Across ten steps, it falls to 60%. The system does not fail loudly or all at once. It silently accumulates latent fragility until a triggering event — a pandemic, a cyberattack, an equipment failure — reveals the structural rot that satisficing methodically installed.

History’s most consequential advances were built on the explicit rejection of this logic.

In the 1950s and 1960s, the semiconductor industry could have satisficed at the transistor — a workable, good-enough technology. Instead, engineers pursued relentless optimization throughout decades, compressing computing power which once filled an entire room into chips that now fit inside a wristwatch, serving billions at a fraction of the first cost. That is not satisficing at scale. That is the compounding dividend of refusing to accept “good enough.”

The biological sciences offer an even starker illustration. In the mid-twentieth century, Malthusian projections of widespread famine appeared mathematically inevitable. Norman Borlaug, the agricultural scientist whose work ultimately saved over one billion lives, did not satisfice with existing crop yield baselines. He pursued grueling, multi-decade optimization of wheat varieties—developing high-yield, disease-resistant strains that triggered the Green Revolution. The counterfactual is a humanitarian catastrophe. The margin between satisficing and optimizing, in this case, was measured in lives at a civilizational scale.

The legendary scientist and industrialist Dr. Arnold O. Beckman understood this intuitively. His foundational corporate maxim—“There is no substitute for excellence”—was not motivational rhetoric. It was an engineering philosophy born from the recognition that in precision instrumentation and scientific research, the accumulation of “good enough” tolerances eventually produces instruments that lie, experiments that mislead, and conclusions that kill.

Today, across a range of existential-related risk domains, the luxury of “good enough” has expired entirely. Industry has long worshiped Six Sigma — a methodology ensuring 99.99966% of outcomes are defect-free. For manufacturing a household appliance, Six Sigma is admirable. But the relevant question for our present technological moment is not whether Six Sigma is better than satisficing. It is whether any standard short of near-absolute certainty is acceptable when the downside is irreversible civilizational harm.

For these domains, I propose what I call the Thirteen-Sigma Standard. In statistical terms, a thirteen-sigma threshold reduces the probability of failure to roughly one in one hundred trillion—a risk so vanishingly small that it effectively ceases to exist over the functional lifespan of the universe. This is not an abstract aspiration. It is a mathematically defined requirement for two categories of existential-level risk: synthetic biological agents, where a single containment failure can propagate across global populations before detection; and autonomous artificial intelligence systems, where an independent self-trained model released without validated safety architecture can trigger domino failures across linked digital infrastructure. In both cases, there is no remediation window. There is no tolerable error rate. The Thirteen-Sigma Standard is not perfectionism  —it is the minimum viable protection for irreversible risk.

Epstein is correct that the tireless pursuit of absolute optimization demands cognitive energy and institutional discipline. But he fundamentally miscalculates the long-term return on that investment. In low-stakes, reversible environments, satisfice freely. Choose your breakfast by whatever heuristic costs you the least. But we do not secure civilization by calibrating our standards to the cognitive limits of our bureaucracies. In the critical architectures—the systems where failure cascades, where errors compound, where the downside is permanent — the obsessive pursuit of optimization is not irrational. It is the only rational choice we have left.