A Fed note reports AI adoption at 18 percent, 78 percent and 41 percent of the same economy

The column separates the statistic from the justification. Pundhir writes that nobody approves a three-year AI program because a survey reports peer adoption, and that the business case carries the justification ^[18]. "What the statistic does is start a clock," he wrote ^[10]. Once a plan opens on peers who have already moved, the argument turns from whether to fund into how fast. He lists what speed buys: compressed vendor selection, diligence deferred to phase two, a pilot promoted before its own evaluation finished ^[11]. All three stay off the line items.

The three readings span 60 percentage points, from 18 to 78, with the household measure of about 41 percent sitting between them ^[1]. The distance is a choice of denominator. One instrument counts firms, and most American firms are small; another weights by employment, and large firms adopt earliest and most ^[5]. Pundhir's point is that the same month and the same government support both a flattering reading and an unflattering one ^[20].

The sharper problem sits in the objective. Take his example. A board sets "reach peer adoption within eighteen months" using a survey where a firm counts as an adopter if any function anywhere uses AI at all ^[12]. Its own reporting counts a function adopted only once a system is deployed, integrated and measured ^[13]. A company can clear the survey bar in a quarter and spend three years clearing its own ^[14]. Three years is twelve quarters, so the internal definition takes roughly twelve times as long for the same word ^[2], and the work lands about eighteen months after the deadline it was set against ^[3].

The Census evidence is the harder part of the case, because behavior held still while the wording moved. Until late 2025 the Business Trends and Outlook Survey asked firms about AI used "in producing goods or services"; on November 17, 2025, that became "in any of its business functions" ^[7]. Cognitive testing had found firms skipping the old question because they did not see themselves as producers, and others answering no before describing AI in hiring, accounting or project management later in the same interview ^[8]. The Bureau then declared the old and new figures incomparable and started a new series on December 4, 2025 ^[9].

"Neither number is wrong. They count different populations," Pundhir wrote ^[6]. His complaint is about practice, including his own. He discloses that in March he used industry adoption and profit-impact figures as peer context and did not say which population they counted, and that he has told executives to bring that kind of benchmark into a board conversation ^[16]. "The figures were real, and the argument holds. The practice was sloppy, and it is the practice this piece is about," he wrote ^[15].

This is one contributor column by the founder of an AI advisory site ^[17]. It identifies the source only as an April 2026 FEDS Note from the Federal Reserve Board and leaves the three instruments it compiles unnamed ^[19]. For a board approving a program this quarter, the binding choice is which definition of "adopted" goes into the objective. Internal reporting will be measured against that definition for the next three years ^[14].