Sometime in the next decade, a human will make the last consequential financial decision. People will still click buttons and sign papers, but the areas where a human's judgment was better than machine's are becoming fewer and further between.
The first wave of financial AI proved both the promise and the limits. Machine learning now dominates credit underwriting. AI-native lenders like Upstart approve 27% more borrowers at 16% lower rates.1 Neural networks run fraud detection. PayPal cut fraud below 0.3%.2 Algorithmic trading accounts for 60-75% of US equity volume.3.
But traditional ML has a fatal flaw: it's fragile. Models trained on historical patterns degrade as markets evolve. Researchers call this concept drift.4 The quant funds that crashed in August 2007 held similar positions because they'd learned from the same history.5 Knight Capital lost $440 million in 45 minutes when its algorithms hit conditions they weren't built for.6 Long-Term Capital Management had Nobel laureates and still blew a $4.6 billion hole in the system.7
This is why traditional ML needs humans today. Not because humans are smarter; in data-rich domains, they're not. But ML systems can't handle the unprecedented. They ask: have I seen this before? When the answer is no, they fail. Humans stayed in the loop as a backstop against novelty.
Large language models change this equation. They reason across domains, interpret ambiguity, and adapt to contexts they've never explicitly seen. Where traditional ML pattern-matches, LLMs understand. They ask: what is this and what should I do?
Production systems have largely solved problems like hallucination through context engineering, domain-specific fine-tuning, and multi-model consensus architectures.8 Morgan Stanley deployed GPT-4 to 16,000 wealth advisors.9 JPMorgan's LLM suite now serves 60,000+ employees.10 Bloomberg built a 50-billion parameter model trained on 363 billion financial tokens.11 Financial LLMs are quickly moving beyond experimentation.
The implications are profound. Traditional ML automated the routine and left humans to handle exceptions. LLMs can handle these exceptions too, but faster and at a greater scale. They read earnings calls, parse regulatory filings, synthesize market sentiment, and can share their reasoning across myriad systems and stakeholders within seconds. The constraints separating human and machine intelligence are dissolving.
Agentic AI accelerates this further. Gartner predicts 15% of day-to-day work decisions will be made autonomously by AI agents by 2028, up from zero in 2024.12 One financial services firm went from 60 autonomous agents in production to over 200 in under a year.13 Citi Research predicts Agentic AI will have a bigger impact on finance than the internet.14
The risks are real but manageable. Concentration could create correlated failures, what researchers call "risk monoculture."15 Speed could amplify crises.16 But these are engineering problems, not fundamental limits. The same was said about algorithmic trading, and we developed circuit breakers and safeguards. Financial AI will evolve the same way. Through deployment, learning, and iteration.
Every platform shift creates new giants. The PC era created Microsoft. The internet created Google. Mobile created Apple's modern incarnation. Finance has resisted. The same names dominate now that dominated thirty years ago. AI breaks that. The firms built in the next few years will shape capital flows for the next fifty.
Run the story forward. AI-mediated finance could be genuinely safer. Risk priced right. Bubbles spotted early. Crises contained. Human judgment gave us 2008.17 The experts missed the risks that destroyed trillions in savings. AI systems that learn from every decision, improve continuously, and operate without ego or fatigue offer something different. Not perfection, but systematic improvement at scale.
The race isn't to build the best model. It's to build the machine that builds the best models. And to deploy it before incumbents wake up.
The firms building now will shape capital flows for decades. The rest will be their customers.