Self-Fulfilling Expectations: How Beliefs Make Unrealized Things Come True
When enough people believe something, it need not come true—its consequences already occur by changing present decisions.
Starting from a quote by Balaji Srinivasan, the article explains "self-fulfilling expectations": beliefs produce real consequences because they change upfront decisions—companies adjust hiring, investors price in, individuals transition early; consequences happen now while facts may never arrive. Bank runs and crypto narratives are classic examples where consensus itself is a factor of production. In the AI era, social media compresses narrative cycles, making self-fulfilling expectations faster, denser, harder to falsify, and solidified by path dependence. But self-fulfilling expectations do not equal realization; narrative-first leads to resource misallocation and product lag. The author advises treating public belief as a real variable and tracking hard signals like capability evaluations, user retention, and cost curves.
When a Belief Becomes Popular Enough, It Doesn't Need to Come True to Change the World
Balaji Srinivasan, Silicon Valley investor and former CTO of Coinbase, once wrote on Twitter: when enough people believe something will happen, it doesn't need to actually happen for its consequences to already be felt.
This sentence may sound convoluted, but it precisely describes a repeatedly validated social and technological phenomenon—self-fulfilling expectations. In financial markets, it's called a "self-fulfilling prophecy"; in the tech industry, "narrative-driven adoption"; in game theory, equilibrium selection in coordination games. And in the AI era, the intensity of this phenomenon is being amplified.
The Transmission Mechanism of Expectations: Why "Not Happening" Still Has Consequences
The Chain Reaction from Belief to Action
The key reason expectations can produce real consequences is that they change people's upfront behavior. If enough people believe that a certain AI model will replace a category of jobs within six months, companies will adjust hiring plans in advance, educational institutions will adjust curricula, and individuals will transition early. These actions themselves will lead to actual changes in job demand—even if that model ultimately never reaches the expected capability.
In other words, beliefs influence present decisions, not future facts. Consequences happen now, while the fact may never arrive.
A Classic Example in Financial Markets
This logic has long had mature examples in finance. A bank run doesn't require the bank to actually be insolvent—as long as enough depositors believe it will collapse, the run itself is enough to bring it down. The same applies to narrative-driven crypto markets: if a protocol is widely believed to gain adoption, capital and developers will rush in ahead of time, thereby genuinely driving adoption.
Balaji himself has long been active in crypto and tech investing, and his observation has its context—in markets highly dependent on consensus, narrative itself is a factor of production.
Coordination game is a basic model in game theory: multiple participants each choose a strategy, and the payoff is maximized when everyone chooses the same strategy, but no one can unilaterally force others to follow. A bank run is a typical coordination game—each depositor's optimal choice depends on their judgment of whether others will withdraw money, and "everyone withdrawing" precisely constitutes a self-fulfilling equilibrium. The diffusion of technology narratives similarly has a coordination game structure: whether developers invest in a platform and whether enterprises adopt a standard both depend on expectations of others' behavior. Once enough people coordinate on the same expectation, that expectation gains stability akin to an "equilibrium," and reversing it requires coordination costs. This also explains why technology narratives, once they reach scale, are hard to shake with individual counterexamples.
The AI Era: Expectations Compressed and Accelerated
Narrowing Time Windows
In the past, the fermentation cycle of a technology narrative could be years, but today the speed of social media allows expectations to reach scale within weeks. When a certain capability (such as "general agents") is widely believed to be imminent, startups will lay out products around it in advance, investors will price it in advance, and regulators will discuss frameworks in advance—all these actions are paying costs or capturing dividends for "something that hasn't happened yet."
This makes self-fulfilling expectations faster, denser, and harder to falsify, because the organizational and capital commitments brought by early action create path dependence.
Path dependence refers to: early choices lock in subsequent development directions through sunk costs, organizational inertia, and the establishment of supporting ecosystems, making it difficult to switch even if better options later emerge. In technology expectations, path dependence manifests as—companies build teams, procure tools, and sign contracts in advance around a narrative that has yet to be delivered; these investments themselves constitute reasons to continue believing in the narrative, because pivoting means admitting prior costs are void. This gives expectations a self-reinforcing nature: not only do beliefs drive action, but action in turn reinforces belief, forming a positive feedback loop. For this reason, the difficulty of falsifying a narrative rises with the scale of investment, and correction often occurs only after costs become explicit.
The Mismatch Between Expectation Bubbles and Real Progress
It is important to be clear: self-fulfilling expectations do not equal expectations coming true. Many technology expectations believed to happen ultimately only changed resource allocation without delivering the capability itself. The result is an industry mismatch of "narrative first, product lagging," with investment and talent drawn to the wrong directions while genuinely promising paths are squeezed out.
Therefore, when judging an AI narrative, an important question is not "will it happen?" but "what are people already doing after believing it?" The latter is the true source of real consequences.
The mismatch of narrative first, product lagging corresponds economically to "resource misallocation": capital and talent are directed toward directions overestimated by narrative, while paths with genuine long-term value are crowded out for lack of attention. A practical way to identify this mismatch is to distinguish "capability signals" from "attitude signals"—the former are reproducible benchmark scores, stable performance in production environments, and quantifiable unit cost reductions; the latter are funding announcements, social media hype, and industry declarations. Attitude signals are easy to manufacture and amplify in the short term, while capability signals are constrained by physics and engineering and cannot be rushed. For practitioners, continuously asking how many reproducible capability signals lie behind a narrative is key to avoiding being swept up in the mismatch.
Implications for Practitioners
Treat Expectations as a Variable, Not Noise
For product and investment decision-makers, public belief is not negligible "sentiment" but a real variable that affects supply and demand, pricing, and talent flows. Understanding the speed and direction of consensus formation is itself a competitive advantage.
Stay Sensitive to Falsification Signals
Since consequences come from beliefs rather than facts, when beliefs shift, consequences also arrive early—and often in reverse. Continuously tracking hard signals that can falsify a narrative (actual capability evaluations, real user retention, cost curves) is more important than chasing the narrative itself.
The reason Balaji's sentence is worth quoting repeatedly is that it reminds us: in consensus-driven systems, the relationship between facts and consequences is not simple causality—there is a layer of crowd belief in between. Understanding how this layer of belief operates is more practically meaningful than waiting for facts to arrive.
One-line Summary
When enough people believe something, it doesn't need to come true—its consequences already happen by changing present decisions.
Paragraph Summary
Starting from a sentence by Balaji Srinivasan, the article explains the phenomenon of "self-fulfilling expectations": beliefs produce real consequences because they change people's upfront decisions—companies adjust hiring in advance, investors price in advance, individuals transition in advance; consequences happen now, while the fact may never arrive. Bank runs and crypto narratives are classic examples, and consensus itself is a factor of production. In the AI era, social media compresses narrative fermentation cycles, making self-fulfilling expectations faster, denser, and harder to falsify, and solidifying them through path dependence. But self-fulfilling expectations do not equal expectations coming true; narrative-first easily leads to resource misallocation and product lag. The author advises practitioners to treat public belief as a real variable affecting supply, demand, and pricing, and to continuously track hard signals such as capability evaluations, user retention, and cost curves, focusing on what people are already doing after believing.
Related articles
Anthropic Files S-1: $16.3B Revenue, $80.7B Net Loss in Half a Year — Who Controls the Company?
Anthropic files S-1 for a Nasdaq IPO: H1 2026 revenue of $16.3B, an $80.7B net loss from valuation remeasurement rather than cash burn, founders' Class F shares holding 50.1% voting power, and no price or date yet.
Anthropic Files S-1: The Truth Behind $16.3B Revenue and $80.7B Loss
Anthropic's S-1 reveals $16.3B revenue and $80.7B net loss in H1 2026, with $72.4B from accounting effects. The filing also discloses $265.6B compute commitments, cloud deal details, and a 50.1% founder voting control structure, while IPO pricing remains blank.
Anthropic Financials Leaked: Revenue Far Exceeds Compute Costs, Claude Inference Gross Margin Hits 68%
Leaked Anthropic financials show revenue roughly twice total compute costs, Claude inference gross margin at 68%, while frontier training burns $2.8B per quarter. This article breaks down the two accounts and analyzes shifts in AI company valuation logic.