The Signal Beneath the Hysteria.

The 2026 AI data is telling three stories at once. The headlines are reading one of them. The Spring Quarterly reads all three, then translates the result into moves you can authorize this quarter and, if you are a parent, into a one-page rubric you can hold across a kitchen table.

The Associated Press ran a story this past Monday with a headline that landed in roughly every parent's inbox in America. Seventy percent of college students now say AI is a threat to their job prospects. The headline ran with the line that no one knows what an AI-proof major actually is. The same week, McKinsey laid off two hundred of its own consultants. CEOs at Ford, Amazon, Salesforce, and JP Morgan publicly forecast that many white-collar jobs will disappear. Two essays comparing this moment to the early days of a serious economic correction were viewed nearly ninety million times between them.

And yet. NBER surveyed six thousand executives and ninety percent reported no AI impact on their firms. Yale Budget Lab, looking at the same months, found unemployment in AI-exposed occupations is essentially flat. Brynjolfsson called the 2025 productivity number, 2.7 percent, nearly double the prior decade's average and the strongest sign yet that the AI productivity take-off is finally visible.

All of these readings are defensible on their own data. They are also incompatible. That incompatibility is itself the signal. Foresight is the discipline that holds up under it.

Five patterns surfaced from the corpus. Three earned promotion to the status of a structural anomaly, meaning a finding that lands in two or more signal classes and shows up in two or more research domains at once. The remaining two stay in monitoring as discount factors applied to the data. The three anomalies are the heart of the issue.

A scenario you have rehearsed lands differently than one you have only read about. That is the operating premise of threatcasting, the discipline Brian David Johnson developed at the Threatcasting Lab at Arizona State University. The work builds futures concrete enough to walk through. Each one becomes a person in a place doing a specific thing on a specific date. The reader does not consume the scenario. The reader inhabits it. By the time the actual moment arrives, the response is already familiar muscle, not improvisation under pressure.

The Spring Quarterly applies the discipline to four protagonists, each sitting at a different seam in the same cascade. The cascade itself is shaped by three forces working in combination. AI capital allocation. Recent energy shocks now modeled as binding constraints rather than variable inputs. The collision between federal preemption and state AI law. The four scenarios are not predictions. They are rehearsal spaces, designed so the reader knows what each future asks of them before the future arrives.

A single excerpt from the Priya threat model gives the texture. Priya is hosting a dinner party for nine people. Her wife Elena is on her second glass of wine. Aditi, her closest peer in the profession, is telling a story about an industry conference panel that ran long. Priya is laughing. The scallops were correctly seared.

The full Priya narrative runs through her seventy-two hours, the regulatory cascade her General Counsel uncovers at midnight, the deployment-over-objection pattern that produced the system in its current form, and the firm's eleven-week remediation that costs eleven million dollars in New York DFS penalties and a separate four point seven million euro EU AI Act fine. None of it is prediction. All of it is rehearsal. The point is not that Priya specifically will be the CISO who absorbs this. The point is that the CISO seat at any regulated firm running agentic AI in production sits on a stack of dependencies, any one of which can trigger a seventy-two-hour crisis.

Read all four narratives in sequence and the cascade is visible in a way no single threat model can show. Rachel's seam tears in September 2027. Kim's tears in June 2028. Priya's tears earliest, in November 2026, because governance is the leading edge of the cascade rather than a lagging one. David's tears last, in March 2029, when the displaced workforce produces a smaller tax base and the municipal layer absorbs the cascade as a structural fiscal squeeze. Four people. Four points. One cascade.

The same logic the Quarterly applies to enterprise capex decisions ports cleanly to the question parents are actually asking right now. Whether to take on a hundred and eighty thousand dollars of debt for a degree whose entry-level job market is contracting in real time. The Family Career Bet is a one-page rubric for that conversation. Two axes. Four quadrants. Three life stage strips. The data anchors are explicit. BLS 2024 to 2034 projections, Tufts AI Jobs Risk Index, Anthropic Economic Index for the cohort effect.

The Quarterly closes with six moves designed to be robust across all four threat models. They are positions designed to hold up regardless of which scenario materializes. Each one is paired with a signal registry of twenty-four indicators, with owners, cadences, thresholds, and pre-named gate responses, so that when a signal fires the response is the decision and the meeting that follows is the execution.

A few examples to indicate the shape. Make the next twenty-four months a monitoring period, and let Q3 NBER, August EU enforcement, and Q4 Yale Budget Lab resolve uncertainty that is currently asking to be resolved by guesswork. Close the governance gap on every agentic workflow with PRISM hard gates on D6, D7, and D8. Redesign the rollout so the incentive structure changes, since forty-four percent of Gen Z is currently sabotaging AI rollouts because their incentives say they should. Stress-test AI unit economics against vendor price resets of fifty to sixty percent in the next twelve to eighteen months. The remaining three live in the issue.