AI and the US Workforce: The Beige Book Tells a Productivity Story

The Federal Reserve’s July Beige Book catches AI arriving in the American labour market as investment: firms buying the tools, reallocating the tasks and holding the headcount. Jamie Dimon’s 30 to 40 per cent is the counterweight that reading must carry, and so far it can.

On 14 July, on JPMorgan’s second-quarter earnings call, Jamie Dimon gave the AI-and-jobs debate its bluntest data point of the year. “We have had discrete areas where we did reduce jobs by 30% or 40%,” the chief executive told analysts, in a quarter for which the bank reported net income of $21.2 billion. The Federal Reserve’s Beige Book appeared the following afternoon. Its material comes from business contacts across all twelve districts, collected on or before 6 July. Anyone expecting the two documents to argue with each other finds something stranger: they describe the same economy, and in that economy AI is not yet a layoff story. It is a capital-spending story.

Photo: Vitaly Gariev / Pexels

Twelve Districts, One Theme

Economic activity increased at a slight to moderate pace in eleven of the twelve districts in late May and June, the national summary reports, and employment rose on balance, with seven districts seeing little to no change. Where AI enters the record, it enters on the investment side of the ledger. A few districts noted firms increasing their use of AI “either in the hiring and screening of potential employees or to boost worker productivity”; the screening half of that sentence extends a shift CFI.co examined in June in The Ghost in the Hiring Machine. The summary’s most telling line is its flattest: “Employers held head counts steady and invested further in AI.”

The district detail sharpens the theme. San Francisco, the district that houses most of the builders, reported employment levels “largely unchanged”, fewer announced or planned layoffs than in prior reporting periods, and firms continuing to invest in AI technologies “seeking to drive productivity improvements”. Some employers there now assess “employees’ ability and willingness to integrate AI agents into the workflow”, a test of who can work alongside the new capital. Atlanta described AI use broadening across its contacts, who are deploying tools and automation “to boost employee productivity and efficiency”; most of those contacts, the district added, “do not expect these efforts to lead to significant workforce reductions in the near term”.

And the scarce worker of mid-2026 is not the prompt engineer. Skilled workers were “difficult to find in a range of fields, notably technicians and tradespeople”, the summary noted, with some wage increases attributed to competition for them; San Francisco contacts described paying “a premium” for specialised, hard-to-fill roles in financial services.

Capital Deepening, in Real Time

At this stage of diffusion, AI spending behaves like earlier rounds of capital deepening: more capital per worker, output per hour rising before any position disappears. That is precisely what the anecdotes describe. Firms buy the tools, reorganise tasks around them and test which employees can supervise them. Headcount waits. The wage premium for technicians and tradespeople is the signature detail, because capital deepening pays the people who install, integrate and maintain the new capital before it pays anyone else. Displacement, where the Beige Book flags it at all, sits in the future tense: Atlanta’s contacts expect no significant AI-driven reductions “in the near term”. That describes intentions. Outcomes arrive later.

The reading has limits. It applies to large US employers at the current stage of adoption, while budgets are fresh and integration skills scarce; it says much less about entry-level white-collar hiring, outsourced back-office work, or the stage at which the agents stop needing supervisors.

The Dimon Counterweight

Any benign account of these anecdotes has to travel with what the country’s most prominent banker said the day before the districts’ reports appeared. Dimon’s reductions of 30 or 40 per cent in “discrete areas” have already happened, and JPMorgan is the kind of early, deep adopter most Beige Book contacts have yet to become. Yet his own gloss pulls the remark back towards the districts’ story. “Most of those people were offered jobs elsewhere,” he said. On margins he was blunter still: “You don’t uniquely benefit from AI,” he argued, since every bank will deploy the same tools for its customers. His chief financial officer, Jeremy Barnum, added that the bank’s generative-AI token expenses, “trivial” today, would see “some meaningful acceleration” in the second half. Heavy investment, tasks reorganised, people redeployed, benefits competed away: the most advanced adopter in American finance is describing capital deepening from further along the same curve.

What the Hard Series Say

The Beige Book is anecdote by design, so the benign reading has to survive contact with the aggregates. Job openings held at 7.6 million in May, the Bureau of Labor Statistics (BLS) reported on 30 June, with layoffs and discharges unchanged at 1.7 million and hires subdued at 5.2 million: a low-hiring, low-firing labour market. Productivity is similarly undramatic. Nonfarm business output per hour grew at an annualised 0.3 per cent in the first quarter, per the BLS release of 4 June. Measured against the same quarter a year earlier, it was up 2.8 per cent. The asymmetry matters: the displacement story has no aggregate evidence at all, while the productivity story has a labour market that keeps absorbing AI investment without shedding workers. In the aggregate data, the layoff wave has not happened yet.

Reasons to Distrust the Comfort

Four cautions keep the reading provisional. First, contacts are telling their Reserve Bank what they intend, and forecasts of one’s own restraint deserve the usual discount. Second, Atlanta’s wording contains displacement’s quiet form: employment “flat to slightly down” as firms hold headcounts even “or adjust downward through attrition and minimal backfilling of roles”. A role never backfilled appears in no layoff count. Third, national openings are too coarse to exonerate anything; 7.6 million vacancies can conceal a collapse in the occupations most exposed to automation. And measured productivity moves for many reasons; crediting the 2.8 per cent to AI would be exactly the inference this piece argues against.

Which Indicator Breaks First

If the benign reading fails, the failure will arrive in a sequence. Occupation-level job postings move first, because attrition with no backfill never registers as a layoff yet shows up straight away in what firms stop advertising; entry-level white-collar postings are the place to stare. The wage distribution moves second: a technology that complements scarce technical skill while substituting for routine cognitive work should widen dispersion before it dents employment. Measured productivity moves last, and noisiest; a genuine boom should eventually hold that 2.8 per cent as hours flatten.

An institutional verdict is also scheduled. On 9 July the Federal Reserve named the leadership of its Productivity and Jobs task force: Marc Andreessen of Andreessen Horowitz, the Stanford growth economist Charles I. Jones, currently on leave at Anthropic, and Microsoft’s Asha Sharma, mandated to “assess the economic impact of new general-purpose technologies, including artificial intelligence, to inform the Federal Reserve’s policy judgments”. Chair Kevin Warsh told his June press conference he hoped most of the task forces, if not all, would conclude “by year-end”, with findings going to the Federal Open Market Committee.

Whether this boom broadens prosperity or concentrates it will be settled in decisions like the one Dimon described: reduce the team, then choose whether its people move elsewhere in the firm or simply out of it. That choice belongs to managements and boards, and the case for treating AI adoption as a governance capability rather than a cost line is one CFI.co made this month in Boards and AI: From Oversight to Insight. The Beige Book’s contacts are, for now, choosing redeployment. Nothing in the document obliges them to keep choosing it.

Sources

1. Federal Reserve, Beige Book, National Summary, July 2026, published 15 July 2026, https://www.federalreserve.gov/monetarypolicy/beigebook202607-summary.htm, accessed 18 July 2026.

2. Federal Reserve, Beige Book, Twelfth District (San Francisco), July 2026, published 15 July 2026, https://www.federalreserve.gov/monetarypolicy/beigebook202607-san-francisco.htm, accessed 18 July 2026.

3. Federal Reserve, Beige Book, Sixth District (Atlanta), July 2026, published 15 July 2026, https://www.federalreserve.gov/monetarypolicy/beigebook202607-atlanta.htm, accessed 18 July 2026.

4. Federal Reserve, press release, “Federal Reserve announces the leadership and objectives of its task forces to advance the conduct of monetary policy”, 9 July 2026, https://www.federalreserve.gov/newsevents/pressreleases/monetary20260709a.htm, accessed 18 July 2026.

5. Federal Reserve, Productivity and Jobs task force page, https://www.federalreserve.gov/monetarypolicy/productivity-and-jobs-task-force.htm, accessed 18 July 2026.

6. Federal Reserve, transcript of Chairman Warsh’s press conference, 17 June 2026, https://www.federalreserve.gov/mediacenter/files/FOMCpresconf20260617.pdf, accessed 18 July 2026 (full text extracted locally).

7. US Bureau of Labor Statistics, Job Openings and Labor Turnover Survey news release, May 2026 data, released 30 June 2026, https://www.bls.gov/news.release/jolts.nr0.htm, accessed 18 July 2026 via search excerpts (direct fetch blocked, see uncertainties).

8. US Bureau of Labor Statistics, Productivity and Costs, First Quarter 2026, Revised, released 4 June 2026, https://www.bls.gov/news.release/prod2.nr0.htm, exact wording verified via the Primary News Source mirror (https://primarynewssource.org/sourcedocument/productivity-and-costs-first-quarter-2026-revised/), accessed 18 July 2026.

9. Business Insider (republished by AOL), “Jamie Dimon says AI already reduced jobs in some areas by 40% … “, 14 July 2026, https://www.aol.com/articles/jamie-dimon-says-ai-already-141858000.html, accessed 18 July 2026.

10. CNBC, “Bank earnings takeaways: From Goldman Sachs’ SpaceX IPO fees to JPMorgan’s AI job cuts”, 14 July 2026, https://www.cnbc.com/2026/07/14/jpm-bank-of-america-citi-bank-earnings-live-updates.html, existence confirmed 18 July 2026; content not retrievable in-session (HTTP 403).

11. Indeed Hiring Lab, “May 2026 JOLTS Report: More of the Same”, 30 June 2026, https://www.hiringlab.org/2026/06/30/may-2026-jolts-report-more-of-the-same/, accessed 18 July 2026.

12. US News (AP), “JPMorgan Chase Profit Hits $16.9 Billion in the Second Quarter, Boosted Again by Market Volatility”, 14 July 2026, https://www.usnews.com/news/business/articles/2026-07-14/jpmorgan-chase-profit-hits-16-9-billion-in-the-second-quarter-boosted-again-by-market-volatility, headline and figure sighted 18 July 2026.

13. CFI.co, “The Ghost in the Hiring Machine: Is AI About to Make the Recruiter Extinct?”, June 2026, https://cfi.co/sustainability/2026/06/the-ghost-in-the-hiring-machine-is-ai-about-to-make-the-recruiter-extinct/, accessed 18 July 2026 (woven per commission).

14. CFI.co, “Boards and AI: From Oversight to Insight”, July 2026, https://cfi.co/northamerica/2026/07/boards-and-ai-from-oversight-to-insight/, accessed 18 July 2026 (woven per commission).