Herbert Simon Won a Nobel Prize for Proving You Can't Make Perfect Decisions. Here's What to Do Instead.

Stockholm, December 8, 1978. Herbert Simon steps up to deliver his Nobel lecture to the assembled economics establishment — and spends it dismantling the model their discipline is built on. The rational agent of the textbooks, who knows all the alternatives, computes all the consequences and selects the maximum? Simon's message, delivered with citations rather than rhetoric: that agent has never existed, and no decision anyone has ever made was produced that way.

The Royal Swedish Academy had already conceded the point. Its press release awarded the prize for pioneering research into the decision-making process within economic organizations and called his 1947 book Administrative Behavior epoch-making — a book whose argument is that organizations are machines for coping with the limits of human reason.

Now run the test on your own organization. Your last vendor selection, your last hiring round, your last strategy call: did anyone enumerate all the alternatives? Compute all the consequences? Of course not — there wasn't time, the information didn't exist, and no head can hold that computation. Simon's point is that this is not a confession of failure. It is the operating condition of every decision you will ever make, and pretending otherwise is what actually damages decision quality.

The Man Who Retired Homo Economicus

Before Simon, the theory of the firm assumed an omniscient, profit-maximizing entrepreneur — homo economicus, equipped with unlimited information, unlimited processing power and unlimited time. Simon replaced that fiction with what he called administrative man: a decision-maker of limited knowledge and limited computation, embedded in an organization, doing the best that boundedness allows.

The replacement matters because the two models give opposite advice. If people optimize, better decisions come from more data, more analysis, more options. If people are bounded, each of those beyond a point makes decisions worse — more data becomes overload, more analysis becomes delay, more options become paralysis. Most decision-improvement programs still quietly assume the first model. Their teams live in the second.

Three Constraints No Amount of Talent Removes

The consequence: humans do not optimize — they simplify. We build a manageable mental model of the decision and act rationally within the model. That is not a bug to be engineered away. It is how intelligent agents necessarily operate, and every serious method for making better decisions starts by accepting it.

Satisficing, in One Paragraph

Simon's name for the rational response is satisficing — satisfy plus suffice. Set an aspiration level (the criteria an acceptable option must meet), search until an option clears it, then stop searching. In his Nobel lecture he put the two available strategies plainly: decision makers can satisfice either by finding optimum solutions for a simplified world, or by finding satisfactory solutions for a more realistic world. What no decision-maker gets is the third option the textbooks assumed — optimum solutions for the real world.

Satisficing is the half of the theory you can run tomorrow morning, and it has its own evidence base — including the awkward finding that people who insist on maximizing make objectively better choices and feel subjectively worse about them. We've given it a full companion piece: how to satisfice deliberately, and when to maximize instead.

Heuristics Are Not Bugs

Decades before behavioral economics made cognitive bias a best-seller genre, Simon read rules of thumb the other way around: heuristics are adaptations to bounded rationality, not failures of it. Go with the trusted vendor, escalate if the risk crosses a threshold, reuse what worked — these compress experience into decisions that are fast, cheap and usually right.

Modern cognitive science has largely come back to Simon's reading. Resource-rational analysis — the framework Falk Lieder and Tom Griffiths consolidated in a landmark 2020 review — models human cognition as the optimal use of limited computational resources: given what thinking costs, the shortcut is often the rational algorithm. The practical question for a team is therefore never whether to use heuristics (you will), but whether the ones you use are any good — and that is a process-design question: encode the experience, make the default the right choice, review the rule when it misfires.

Why Organizations Exist at All

Read that list against your own company and the reframe is hard to unsee: structure is cognitive architecture. Which also means bad structure is a cognitive defect — a hierarchy that filters out the wrong signals, routines that pre-compute yesterday's answer, and decision records that were never written are failures of the machine organizations exist to be.

If Perfection Is Impossible, Why Improve Anything?

The objection arrives on schedule: if perfect rationality is unreachable, isn't this a license for mediocre decisions? Simon's answer is the opposite, and it is the most practical thing he wrote. You cannot close the gap between bounded minds and objective rationality by trying harder — but you can move the boundary, because the boundary is set partly by the environment the decision is made in.

The wrong program: demand more data, more analysis, more options, and call the resulting delay rigor. Beyond a modest point each of these degrades the decision — that is bounded rationality operating on the improvement program itself, and it is how teams end up with decision fatigue instead of decision quality.

Simon's program: design better environments for bounded humans. Clearer processes, so ambiguity doesn't consume cognition. Explicit criteria, so search has a stopping rule. Externalized reasoning, so the argument lives outside any one head. Documented archives, so organizational memory compounds. The goal is not maximum quality on one heroic decision — it is sufficient quality at the right speed, reliably, across all of them.

2025 Postscript: The Machines Are Bounded Too

Simon's theory is having a second scientific life, because the newest decision-makers in the building are bounded as well. A 2025 study of strategic decision-making found that large language models deviate from full game-theoretic rationality in recognizably human ways — reproducing familiar heuristics, only applied more rigidly (arXiv:2506.09390). Another 2025 line of work builds satisficing directly into model alignment: maximize a primary objective while holding secondary criteria to acceptable thresholds — Simon's aspiration levels, implemented at inference time (arXiv:2505.23729).

He would not have been surprised. He spent the second half of his career building machines on exactly this principle — that intelligence means smart search under constraints, not exhaustive computation. That story, from the first AI program to the modern echoes, is the third piece in this series. And his most-quoted sentence, written in 1971, reads like a diagnosis of the attention economy filed fifty years early.

The Technique: Design the Environment, Not the Decider

Simon's prescription compresses into five moves a team can adopt this quarter:

Where Argumentree Fits

Argumentree is Simon's environment-design program, built as software. The argument tree externalizes reasoning — every pro, con and piece of evidence visible in one structure instead of scattered across heads and threads. Ratings surface the strongest arguments so attention, the scarce resource, goes where it matters. And the permanent, searchable record turns each decision into organizational memory the next decision can build on.

None of that makes a team perfectly rational — nothing does; that is the theorem. It moves the boundary. Which is exactly what Simon said improvement actually looks like.

The Optimistic Reading

Bounded rationality sounds like a pessimistic result and is the opposite. Simon spent his career showing that the limits are real and that they are workable: with the right structures, bounded minds routinely produce decisions far beyond any individual's cognitive reach. The Nobel lecture is not an elegy for rationality — it is an engineering manual for extending it.

So retire the fantasy decision-maker, and stop budgeting for his arrival. The work is not to become unbounded. The work is to build environments in which bounded is enough.