The cognitive bench is advisory's real AI exposure

Through the end of 2025, roughly 18% of US firms and 41% of US workers reported using AI, with adoption running heaviest among larger employers and in knowledge-intensive sectors such as professional services and finance. Diffusion is no longer the question. What The Conference Board's new research, reported by InvestmentNews, sets out to answer is what all that usage does to jobs and wages, and it does so by declining to issue a single forecast: four scenarios, from gradual augmentation to concentrated gains to massive displacement to uneven disruption. Read from benign to brutal, those paths have AI assisting workers rather than replacing them; a narrow set of industries and roles capturing most of the productivity benefit; job losses on a scale rivaling the most severe economic shocks in US history; and heavy losses in some occupations alongside growth in others. The report's advice to business leaders, policymakers and educators is to prepare now rather than wait for certainty.

Certainty, for now, is not showing up in the data: worker-level productivity gains and employment effects have been slow to materialize and hard to measure, which is why the report reaches for the Solow paradox—Robert Solow's 1987 observation that the computer age was visible everywhere except in the productivity statistics. Those gains eventually proved real, arriving roughly two decades later. The Conference Board suggests AI may follow a similar delayed trajectory, with the qualification that its reach into cognitive work could make the transition faster and more disruptive than the computer's. A firm that cannot see where AI is generating gains cannot pay for them, and the benefit then accrues to whoever captures it in margin or price rather than in the labor line.

Treat the four futures as distribution math rather than HR scenarios. Advice reaches clients through people, and a displacement path on the scale described would shrink the pool of humans available to carry relationships—making it an asset-gathering problem before it is a cost problem. The concentrated-gains path reads gentler and may not be: it would let a small number of firms compound a productivity advantage the rest cannot match, and it is the scenario the coverage singles out as particularly relevant to wealth management.

That advantage has a short clock. Within three years, the Conference Board projects, 60 to 70% of jobs in the cognitive workforce will involve collaboration between humans and AI, against 15 to 25% involving human-only work. Cognitive, in the report's usage, means roles built on knowledge and information tasks rather than manual labor—the job families advisory firms staff, from paraplanners and investment analysts to service associates and relationship managers.

The awkward part is the span. Previous automation waves mostly took middle-skill routine jobs; this one reaches across the income spectrum to high earners and knowledge workers who came out ahead of the last technology cycle, while workers in trades and manual labor may be less directly exposed and could see relative gains. The Conference Board's AI and Automation Risk Index, built on Occupational Information Network data, estimated in 2024 that roughly 48% of tasks in STEM fields carry high task exposure to AI, against 23% of tasks in manual trades and production.

The reflex inside advisory firms is to file AI exposure under back office, where the routine work was supposed to live. The index points at analytical and information-processing tasks instead—what a paraplanner produces all day and what an analyst assembles for every rebalance. If a firm's growth model runs on adding junior staff to absorb client work, that is the model the index puts on notice.

Headcount reduction is the wrong first read for a business whose product is judgment. Firms that cut the bottom of the pyramid without deciding what the pyramid was for will spend the savings recruiting senior people they have no bench to support. The better read is reallocation: the same revenue carried by fewer production seats and more client-facing time. As this publication has argued, AI-for-advisors value is settling with the platforms and custodians that own the connector, with queue position and plumbing mattering ahead of model quality, and the labor research supplies the corollary. If collaboration becomes the default, seats per dollar of revenue turns into a strategy variable, and the connector thesis only pays if the firm has decided how many of those seats it needs.

For a household whose wealth is a knowledge-worker salary plus equity compensation, human capital is the largest asset on the balance sheet, and it now sits in the cohort the risk index scores highest. That argues for a planning question rather than a market call: the income stream funding the plan deserves the same stress test an advisor would apply to a concentrated equity position, and clients whose earnings come from STEM and knowledge-sector work are the ones the report describes as newly exposed.

Three years is a budget cycle, not a forecast horizon, which is why the collaboration range is the number to plan against even while the productivity statistics sit still. If the firm-level answer to AI turns out to be a seat count rather than a model, watch the job postings: the first real evidence that a firm has absorbed this research will be a paraplanner or analyst listing with AI oversight written into the responsibilities and a different pay band underneath it.