arxiv: 2604.19833 · v1 · submitted 2026-04-21 · 💰 econ.EM · cs.CY· physics.ao-ph

Recognition: unknown

From Clerks to Agentic-AI: How will Technology Change Labor Market in Finance?

Lu Yu, Xiang Li

Authors on Pith no claims yet

Pith reviewed 2026-05-10 02:08 UTC · model grok-4.3

classification 💰 econ.EM cs.CYphysics.ao-ph

keywords asset managementtechnology waveslabor productivityAUM per employeeAI automationpassive investingfinancial firmsstylized facts

0 comments

The pith

Technology waves in finance have changed the labor scale for managing assets, shown by tracking AUM per employee across computerization, indexing, and AI periods.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines how three successive technology waves have affected the labor needed to manage capital in financial firms. It uses a small panel of representative firms to compare assets under management per employee, revenue per employee, and operating expense intensity over time. The analysis covers computerization in the 1980s and 1990s, the rise of passive investing in the 2000s and 2010s, and AI and automation from 2015 onward. Rather than establishing causal links, the work documents stylized facts about shifts in the scale of asset management work. A reader would care because these patterns help anticipate how agentic AI might further reshape employment and operations in finance.

Core claim

Using a small panel of representative firms, the authors compare changes in AUM per employee, revenue per employee, and operating expense intensity over time to document stylized facts about how technology changes the scale of asset management work across the computerization, indexing, and AI waves.

What carries the argument

Assets under management per employee, serving as a simple productivity proxy for the labor required to manage capital across technological periods.

If this is right

Each technology wave corresponds to measurable shifts in how much capital an employee can oversee.
Revenue per employee reflects combined effects of technology adoption and operational scaling.
Operating expense intensity can be compared to assess cost efficiency gains across periods.
The AI wave is included in the same descriptive framework as prior waves for consistency in tracking labor scale.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If the observed patterns continue, AI tools could allow even larger asset volumes to be managed with stable or reduced headcount in routine operations.
Finance roles may shift toward oversight and exception-handling rather than direct portfolio work.
Extending the panel to more firms or adding controls for market size could test the robustness of the documented trends.
Similar productivity ratios might be applied to other financial activities like lending or trading to compare tech impacts.

Load-bearing premise

Assets under management per employee serves as a reliable proxy for the labor required to manage capital and the small panel of firms is representative of broader industry trends.

What would settle it

Panel data showing no increase or a decline in AUM per employee during the AI wave since 2015 would indicate that technology has not altered the labor scale in the manner described.

Figures

Figures reproduced from arXiv: 2604.19833 by Lu Yu, Xiang Li.

**Figure 2.** Figure 2: Simple finance productivity eras real The real-era figure reinforces this pattern after deflating the productivity ratios, indicating that the increase is not simply an artifact of nominal growth. B. Automation and Productivity The AI-focused filing analysis examines whether firms with rising AI disclosure intensity also exhibit systematic changes in operating outcomes. The fixed-effects estimates indicat… view at source ↗

**Figure 3.** Figure 3: Coefficient plot of AI exposure fixed-effects results [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Wage trends by group real D. Heterogeneity The event-study style figure suggests that adoption is staggered across firms rather than synchronized at the industry level. Bank of America appears earlier, S&P Global accelerates later, JPMorgan and State Street move more gradually, and BNY Mellon crosses the threshold very late in the sample. This timing pattern points to heterogeneous organizational adoption… view at source ↗

**Figure 5.** Figure 5: Event-study style AI exposure dynamics VII. Transformation for the Financial Labor Market Taken together, the evidence points to transformation rather than simple replacement. Productivity gains appear before labor-cost reductions, suggesting that automation in finance initially operates by changing how work is organized rather than by immediately eliminating jobs. Firms can use technology to expand monito… view at source ↗

read the original abstract

Financial firms have gone through three major technological waves: computerization in the 1980s and 1990s, the rise of indexing and passive investing in the 2000s and 2010s, and the AI and automation wave from roughly 2015 to the present. This project studies how much labor is required to manage capital across those waves by tracking a simple productivity measure: assets under management per employee. Using a small panel of representative firms, we compare changes in AUM per employee, revenue per employee, and operating expense intensity over time. The goal is not to identify causal effects, but to document stylized facts about how technology changes the scale of asset management work.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper documents rising AUM per employee across three finance tech waves using a small panel but the ratio mixes productivity with market price gains.

read the letter

The paper tracks assets under management per employee, revenue per employee, and operating expense intensity in a small panel of finance firms over the computerization era, the indexing period, and the AI wave since 2015. It frames the exercise as stylized facts rather than causal work, which matches what the data can support. The timeline is laid out clearly and the ratios are shown moving in the expected direction of greater scale per worker. This gives a set of concrete historical benchmarks that could anchor talks about AI and headcount in asset management. The approach is a direct application of standard productivity ratios to these specific breaks, so it does not introduce new methods but it does organize the periods in one place. The main weakness is that AUM per employee rises with asset price appreciation even if technology, processes, or labor input stay fixed. The description gives no indication of deflating by a broad market index or checking robustness with other scale variables such as number of accounts or trades. A small panel also leaves the representativeness open without more detail on selection. Readers who want quick descriptive context on technology and finance labor will find the numbers useful as a starting point. It is not the place to look for identification or new theory. The work deserves peer review because the topic is current, the descriptive goal is stated plainly, and referees can push on the measurement and sample issues in revisions.

Referee Report

2 major / 2 minor

Summary. The manuscript documents stylized facts on labor requirements in asset management across three technological waves: computerization (1980s-1990s), indexing and passive investing (2000s-2010s), and AI/automation (2015-present). It tracks changes in assets under management (AUM) per employee, revenue per employee, and operating expense intensity using a small panel of representative firms, with the stated goal of describing trends in the scale of work rather than establishing causal effects.

Significance. If the trends prove robust after addressing measurement confounds and providing sample details, the paper could offer useful descriptive benchmarks for how technology has scaled operations in finance. However, the current lack of data transparency and adjustment for exogenous factors limits its ability to inform labor-market analyses or distinguish technology-driven productivity from market valuation effects.

major comments (2)

Abstract and methods description: AUM per employee is presented as a proxy for labor required to manage capital, but the manuscript does not deflate AUM by a market index (e.g., S&P 500 or broad bond index) or report robustness checks using alternative scale measures such as number of accounts, trades, or regulatory filings. Because AUM equals quantities times prevailing valuations, observed increases could mechanically reflect post-GFC asset-price appreciation rather than technology-induced reductions in labor input.
Methods description: The manuscript supplies no information on the identity or characteristics of the small panel of representative firms, the precise time periods and data sources used, sample size, or any stratification to hold market exposure constant. Without these details, it is impossible to assess whether the panel supports the claim of documenting industry-wide stylized facts or whether results are sensitive to selection.

minor comments (2)

Abstract: The term 'operating expense intensity' is introduced without a definition or formula, making it unclear how this metric is constructed or normalized.
Abstract: The claim that the panel consists of 'representative firms' is stated without supporting evidence or criteria for representativeness.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the scope and limitations of our descriptive analysis. We agree that greater transparency on measurement issues and sample construction will strengthen the manuscript as a set of stylized facts. We address each major comment below and indicate the planned revisions.

read point-by-point responses

Referee: Abstract and methods description: AUM per employee is presented as a proxy for labor required to manage capital, but the manuscript does not deflate AUM by a market index (e.g., S&P 500 or broad bond index) or report robustness checks using alternative scale measures such as number of accounts, trades, or regulatory filings. Because AUM equals quantities times prevailing valuations, observed increases could mechanically reflect post-GFC asset-price appreciation rather than technology-induced reductions in labor input.

Authors: We appreciate this point on measurement. The paper is explicitly descriptive and does not attempt to isolate technology effects from valuation changes. We will revise the methods section to note that AUM reflects both quantities managed and prevailing market prices, and that post-GFC appreciation may contribute to observed trends. We will also add a discussion of revenue per employee as a complementary measure less directly tied to valuations, along with a caveat on the limitations of AUM as a labor-input proxy. Full robustness checks with alternative scales (e.g., accounts or filings) will be noted as data-constrained but discussed where possible. revision: partial
Referee: Methods description: The manuscript supplies no information on the identity or characteristics of the small panel of representative firms, the precise time periods and data sources used, sample size, or any stratification to hold market exposure constant. Without these details, it is impossible to assess whether the panel supports the claim of documenting industry-wide stylized facts or whether results are sensitive to selection.

Authors: We agree that additional sample details are needed for transparency. In the revision we will expand the data section to describe the panel's characteristics (e.g., firm size, asset-class focus), the exact time periods covered, data sources (annual reports and financial statements), sample size, and selection criteria. We will also discuss the panel's market exposure and any limitations on generalizability to the broader industry. revision: yes

Circularity Check

0 steps flagged

No circularity; purely descriptive documentation with no derivation chain

full rationale

The paper states its goal is to document stylized facts about labor requirements in asset management by tracking AUM per employee, revenue per employee, and operating expense intensity across three technological waves using a small panel of firms. It explicitly disclaims causal identification. No equations, models, fitted parameters, predictions, self-citations, or ansatzes appear in the provided text. The central measure is introduced directly as a simple productivity proxy without any reduction to prior inputs or self-referential definitions. The analysis is self-contained empirical observation and contains no load-bearing steps that could be circular by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on treating AUM per employee as a valid labor-productivity proxy and assuming the chosen firms represent the industry.

axioms (1)

domain assumption Assets under management per employee accurately reflects the labor required to manage capital
Invoked as the primary productivity measure without further justification in the abstract.

pith-pipeline@v0.9.0 · 5414 in / 1112 out tokens · 37783 ms · 2026-05-10T02:08:01.693076+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

24 extracted references

[1]

Tasks, Automation, and the Rise in U.S. Wage In- equality

Acemoglu, D., and P. Restrepo. 2022. “Tasks, Automation, and the Rise in U.S. Wage In- equality.” Working paper

2022
[2]

Tasks at Work: Comparative Advantage, Technology and Labor Demand

Acemoglu, D., F. Kong, and P. Restrepo. 2025. “Tasks at Work: Comparative Advantage, Technology and Labor Demand.”Handbook of Labor Economics6: 1–114

2025
[3]

How AI Aggregation Affects Knowledge

Acemoglu, D., T. Lin, A. Ozdaglar, and J. Siderius. 2026. “How AI Aggregation Affects Knowledge.” Working paper

2026
[4]

AI, Human Cognition and Knowledge Collapse

Acemoglu, D., D. Kong, and A. Ozdaglar. 2026. “AI, Human Cognition and Knowledge Collapse.” Working paper

2026
[5]

Automation and Polarization

Acemoglu, D., and J. Loebbing. 2026. “Automation and Polarization.”Journal of Political Economy134(3): 1017–1072

2026
[6]

Building Pro-Worker Artificial Intelligence

Acemoglu, D., D. Autor, and S. Johnson. 2026. “Building Pro-Worker Artificial Intelligence.” Brookings Institution

2026
[7]

Innovation and Top Income Inequality

Aghion, P., U. Akcigit, A. Bergeaud, R. Blundell, and D. H´ emous. 2016. “Innovation and Top Income Inequality.”Journal of Political Economy124(5): 1473–1504

2016
[8]

Bergeaud, T

Aghion, P., A. Bergeaud, T. Boppart, and J.-F. Brouillette. 2025.Resetting the Innovation Clock: Endogenous Growth through Technological Turnover. Book manuscript

2025
[9]

A Research Agenda for the Economics of Transformative AI

Agrawal, A. K., E. Brynjolfsson, and A. Korinek. 2025. “A Research Agenda for the Economics of Transformative AI.” NBER Working Paper

2025
[10]

Agrawal, A. K., E. Brynjolfsson, and A. Korinek. 2025.The Economics of Transformative AI. Chicago: University of Chicago Press

2025
[11]

Does Automation Replace Experts or Augment Expertise? The Answer Is Yes

Autor, D. 2024. “Does Automation Replace Experts or Augment Expertise? The Answer Is Yes.” Schumpeter Lecture, European Economic Association

2024
[12]

Expertise

Autor, D., and N. Thompson. 2025. “Expertise.”Journal of the European Economic Associ- ation23(4): 1203–1271

2025
[13]

Resolving the Automation Paradox: Falling Labor Share, Rising Wages

Autor, D., and B. Kausik. 2026. “Resolving the Automation Paradox: Falling Labor Share, Rising Wages.” Working paper

2026
[14]

What Makes New Work Different from More Work?

Autor, D., C. Chin, A. Salomons, and B. Seegmiller. 2026. “What Makes New Work Different from More Work?” Forthcoming inAnnual Review of Economics

2026
[15]

The Turing Trap: The Promise and Peril of Human-Like Artificial Intelligence

Brynjolfsson, E. 2022. “The Turing Trap: The Promise and Peril of Human-Like Artificial Intelligence.”Daedalus

2022
[16]

Generative AI at Work

Brynjolfsson, E., D. Li, and L. Raymond. 2023. “Generative AI at Work.”Quarterly Journal of Economics. 11

2023
[17]

The Macroeconomics of Artificial Intelligence

Brynjolfsson, E., and G. Unger. 2023. “The Macroeconomics of Artificial Intelligence.” IMF working paper

2023
[18]

AI Adoption in America: Who, What, and Where

Brynjolfsson, E., K. McElheran, J. F. Li, Z. Kroff, E. Dinlersoz, L. Foster, and N. Zolas. 2024. “AI Adoption in America: Who, What, and Where.”Journal of Economics & Management Strategy

2024
[19]

New World Order: Labor, Capital, and Ideas in the Power Law Economy

Brynjolfsson, E., A. McAfee, and M. Spence. 2014. “New World Order: Labor, Capital, and Ideas in the Power Law Economy.”Foreign Affairs, July/August

2014
[20]

Will Humans Go the Way of Horses? Labor in the Second Machine Age

Brynjolfsson, E., and A. McAfee. 2015. “Will Humans Go the Way of Horses? Labor in the Second Machine Age.”Foreign Affairs, July/August

2015
[21]

The Business of Artificial Intelligence: What It Can and Cannot Do for Your Organization

Brynjolfsson, E., and A. McAfee. 2017. “The Business of Artificial Intelligence: What It Can and Cannot Do for Your Organization.”Harvard Business Review

2017
[22]

Minimum Wages and Rise of the Robots

Brynjolfsson, E., J. F. Li, J. Miranda, R. Seamans, and A. J. Wang. 2026. “Minimum Wages and Rise of the Robots.” NBER Working Paper 34895

2026
[23]

Post-AGI Economics As If Nothing Ever Happens

Kulveit, J. 2026. “Post-AGI Economics As If Nothing Ever Happens.”Less- Wrong, February 4, 2026.https://www.lesswrong.com/posts/fL7g3fuMQLssbHd6Y/ post-agi-economics-as-if-nothing-ever-happens#

2026
[24]

Human Work in the Robotic Future: Policy for the Age of Automation

McAfee, A., and E. Brynjolfsson. 2016. “Human Work in the Robotic Future: Policy for the Age of Automation.”Foreign Affairs, July/August. 12

2016