From Clerks to Agentic AI: A 40-Year Productivity Trend Says Finance Jobs Are Restructuring, Not Disappearing

A new arXiv paper just put a number on something the finance industry has been quietly experiencing for forty years — and it is not the number you would expect.

[Claim] When most people think about AI replacing finance jobs, they picture trading floors emptying out next year. The actual story, according to a paper posted to arXiv on April 21, 2026, is slower, longer, and already half-finished. The researchers tracked one stubborn metric across three technological waves — and what it shows about the future of finance work might surprise you.

The metric is assets under management per employee. And it has been climbing since the 1980s.

The Paper, In Plain English

The paper is titled "From Clerks to Agentic-AI: How will Technology Change Labor Market in Finance?" by Lu Yu and Xiang Li, posted to arXiv as paper 2604.19833.

The thesis is deceptively simple. Financial firms have been through three technology waves: [Fact] 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. Each wave reshaped how much human labor was required to manage capital. The authors track that change with a single productivity ratio — assets under management per employee — using a small panel of representative firms.

They are clear about their limits. "The goal is not to identify causal effects, but to document stylized facts about how technology changes the scale of asset management work." [Claim] Translation: this is not a prediction model. It is a long-exposure photograph of an industry slowly automating itself, with the labor floor rising one decade at a time.

Why The AUM-Per-Employee Number Tells The Story

Assets under management per employee sounds like accountant jargon. It is actually one of the cleanest single-number summaries of how much technology a firm has absorbed.

In the 1980s, processing trades, reconciling positions, and reporting to clients required armies of clerks. [Estimate] A typical asset manager in 1985 needed dozens of back-office staff per billion dollars under management. By the 2010s, after computerization and indexing, that ratio had inverted dramatically — the same billion dollars could be managed with a fraction of the headcount.

The Yu and Li paper extends that arc into the AI era. [Claim] Their argument, supported by their panel data, is that the AI and automation wave from 2015 onward has continued the trend at roughly the same trajectory — not the cliff-edge displacement that some forecasters predicted, but the same steady rise in assets per employee that the industry has produced for forty years.

That distinction matters. A cliff means one bad year for finance workers. A continued slope means a gradual, decade-long restructuring of who works in the industry and what they do.

What The U.S. Bureau Of Labor Statistics Says

The Yu and Li paper does not include forward-looking employment forecasts. But the U.S. Bureau of Labor Statistics has published projections that line up with their stylized-facts story rather than the cliff-edge story.

According to the BLS Occupational Outlook Handbook for the 2024-2034 projection period, [Fact] personal financial advisors — the largest occupation in the finance industry — are projected to grow 9.6 percent over the decade, much faster than the average for all occupations. Financial and investment analysts are projected to grow 5.7 percent, and securities, commodities, and financial services sales agents are projected to grow 3.3 percent.

[Claim] Those are not the numbers of an industry being eliminated. They are the numbers of an industry restructuring. The BLS commentary is explicit about why: AI-powered robo-advisors have emerged as alternatives to human advisors, but "this will only have a mild effect on employment for these workers, as older clients with sophisticated financial planning needs are unlikely to trust automated recommendations."

The Yu-Li paper and the BLS projections agree on the underlying story. AUM per employee keeps rising. Headcount in advisory and analyst roles keeps growing modestly. The math works because the industry is also growing — more capital flows in than the productivity gains absorb.

Who Wins And Who Loses Inside Finance

Reading the Yu-Li paper alongside the BLS data, three patterns emerge for finance workers.

[Estimate] Pattern one: client-facing advisory roles look durable. Personal financial advisors growing 9.6 percent over the decade is the strongest signal in the dataset. The reason — trust in human judgment for complex financial planning — is exactly the kind of capability that AI systems have not closed the gap on.

Pattern two: analyst roles get reshaped, not eliminated. Financial and investment analysts growing 5.7 percent suggests that the work is being augmented, not automated away. [Claim] If you are an analyst today, the realistic threat is not unemployment in 2030 — it is needing to operate at a higher abstraction level, with AI handling pattern-finding while you handle judgment, narrative, and client communication.

Pattern three: pure back-office processing roles continue their forty-year decline. The Yu-Li paper does not break out specific occupations, but the trend it documents — AUM per employee rising every decade — is mechanically driven by the disappearance of clerical processing work. [Estimate] If your job is reconciliation, position-keeping, or routine reporting, the rise of agentic AI represents a continuation of pressures that started with mainframes in 1985.

What The "Agentic AI" In The Title Actually Means

The paper's title puts "Agentic-AI" in the same sentence as 1980s clerks for a reason. [Claim] Agentic AI — systems that can plan, take actions, and complete multi-step workflows autonomously — represents the technology category most likely to extend the AUM-per-employee trend into a new productivity regime.

The first computerization wave automated calculations. The indexing wave automated portfolio construction. Agentic AI, if the technology delivers on its premise, automates the workflow itself: not just the trade execution, but the research, drafting, monitoring, and reporting around it.

The Yu-Li paper does not quantify how big that next jump will be. But the implicit framing is significant: they are placing agentic AI in the same lineage as the prior waves, not treating it as a sudden departure. [Estimate] If they are right, the financial industry's labor restructuring continues at roughly its forty-year pace — slower than dramatic predictions, faster than stability.

The Bottom Line For Finance Workers

If you work in finance today, the Yu-Li paper offers a calmer reading of your future than most AI labor coverage suggests.

The industry will keep absorbing technology. Assets per employee will keep rising. Some roles — back-office, processing, reconciliation — will keep shrinking, just as they have for four decades. Other roles — financial advisor, analyst, client-facing advisory — will keep growing modestly because human trust and judgment remain the hardest things to automate.

The honest summary is that finance is not facing an extinction event in 2026. It is facing the next chapter of a transition that has been running since the introduction of mainframe trading systems in the 1980s. [Claim] Workers who treat agentic AI as a new tool to learn, the way prior generations learned spreadsheets and Bloomberg terminals, are likely to do well. Workers who treat it as something happening to other people are not.

---

This article was written with AI assistance. The primary source is the arXiv paper "From Clerks to Agentic-AI: How will Technology Change Labor Market in Finance?" by Lu Yu and Xiang Li (arXiv:2604.19833, April 21, 2026). Employment projections are sourced from the U.S. Bureau of Labor Statistics Occupational Outlook Handbook for 2024-2034.