Finance Is Dismantling Its Own Training Pipeline

OpenAI posted a job listing this summer that does not make sense until you see the loop. The company is recruiting an investment banking professional for its Applied AI team — someone who knows how the work is actually done.

You will bring deep, current knowledge of how investment banking work is actually performed, including company and industry research, financial analysis and modeling, valuation, diligence, transaction execution, and the creation and review of client materials. — OpenAI

The AI lab building the tools that automate finance is hiring the exact people whose expertise those tools encode — from an industry that is, at the same moment, dismantling the entry-level jobs that produce that expertise in the first place. ^[1] The cuts are not a crisis response. They are how the industry is choosing to build its workforce now. Standard Chartered CEO Bill Winters made the logic explicit.

will eliminate jobs. — Jamie Dimon

Jamie Dimon was blunter: some companies, he said, should never have hired those people in the first place. ^[2] Global banks are cutting junior analyst classes by up to two-thirds. ^[2] JPMorgan has deployed agentic AI to automate the creation of pitch decks — the foundational work junior bankers used to learn the craft. ^[3] Standard Chartered is eliminating 7,000 to 8,000 roles by 2030; HSBC is weighing roughly 20,000 administrative cuts. ^[4] Dimon says JPMorgan will hire more AI specialists and fewer traditional bankers, and asks, of the pace, what happens if it moves too fast. ^[4] Meanwhile, the people who did climb that ladder are leaving. Vertical AI startups Rogo and Hebbia are actively recruiting from Goldman Sachs, JPMorgan, and Citi — not a mass exodus, but a significant increase, with roughly 60% of Hebbia's workforce now coming from finance backgrounds. ^[5] Rogo has reached 150 employees and a $2 billion valuation. ^[5] And OpenAI's listing makes explicit what was already underway: the AI labs themselves are now the destination for the domain expertise the banks spent decades cultivating. ^[1] What those former bankers are building is precisely what makes the junior cuts possible. Hebbia's tools, already embedded at KKR, Centerview, and Permira, automate research, financial modeling, and slide decks — the very work junior bankers used to do. ^[6][5] Anthropic has launched specialized Claude plugins for investment banking that integrate directly into Excel, PowerPoint, and Slack — the exact tools where junior analysts once learned to build a model or structure a pitch. ^[7] The expertise that was once built inside the institution, through years of late-night deck revisions and model checks, is now being encoded into software sold back to it. The training ground itself is disappearing. GoTo CEO Rich Veldran calls those foundational tasks the rite of passage that builds intuition and critical thinking — remove them, and you remove the mechanism by which a twenty-two-year-old becomes someone who can evaluate a deal. ^[8]

It’s only in the rearview mirror that folks will look back and say, ‘You know what, I’m actually relying on it too much. Perhaps I’m not learning some of the things I need to learn,' — Rich Veldran

The World Economic Forum warns that eliminating entry-level positions could weaken long-term talent pipelines and create critical capability gaps. ^[9] PwC is cutting graduate hiring; Australian entry-level vacancies have fallen to their lowest in a decade; U.S. college-graduate unemployment has surpassed the national average for the first time in ten years. ^[10][11][12] The counter-evidence is real and worth taking seriously. The European Central Bank finds that AI-intensive firms tend to hire more, not less. ^[13] Stanford's research shows AI has had only a small impact on aggregate employment so far. ^[14] Mizuho is replacing 5,000 administrative jobs with AI but retraining and redeploying those workers rather than laying them off. ^[15] And the doorman fallacy — companies reversing AI-driven layoffs after discovering the humans were doing work the software could not — has played out at Commonwealth Bank of Australia and elsewhere. ^[16] But none of these break the finance-specific loop. The ECB and Stanford findings span sectors and skill levels far beyond the narrow band of entry-level professional-services roles where the cycle is concentrated. Mizuho's retraining program addresses administrative and clerical positions, not the junior analyst track where judgment is forged. And the doorman-fallacy reversals have involved customer service and routine tasks — not the financial modeling and pitch-deck work that the AI tools are now absorbing. The loop does not require every bank to cut every junior role. It only requires enough of them to do so that the pipeline thins, the expertise migrates, and the tools those migrants build make further cuts easier to justify. The result is an inversion of the traditional career arc. The expertise that was once built inside the institution — through years of apprenticeship on the deals that taught the craft — now appreciates in value only outside it, at the companies automating the institution. The banks are paying for the tools; the people who could train the next generation are building them instead; and neither side of the loop has a plan for where the next generation of judgment comes from.