In February 2026, Goldman Sachs CEO David Solomon confirmed what Wall Street had been whispering: AI agents now handle tasks that previously required teams of junior analysts. The bank reduced its incoming analyst class by 30% — not because of hiring freezes, but because AI handles the work.
What the AI Agents Actually Do
Financial modeling: Building DCF models, sensitivity tables, and comparable company analyses. What took a first-year analyst 15 hours now takes an AI agent 8 minutes.
Earnings analysis: Parsing 10-Ks, earnings transcripts, and guidance revisions across hundreds of companies simultaneously. The AI flags anomalies humans consistently miss.
Market research: Synthesizing industry reports, news, and data into investment memos. The output quality matches senior associate-level work.
Due diligence: For M&A deals, AI agents now handle initial target screening, financial health checks, and risk flag identification.
The Numbers Are Staggering
- Cost per analysis: ~$0.50 with AI vs. ~$150 in junior analyst labor
- Speed: 300x faster on routine financial modeling
- Accuracy: 40% fewer errors in data extraction and calculation
- Coverage: One AI agent monitors 5,000+ securities vs. a team covering 50
Which Roles Are Safe (For Now)
Client relationships: No AI is closing a $2B M&A deal over dinner at Nobu. Relationship management remains human.
Creative deal structuring: Novel financial instruments and bespoke solutions require judgment that AI lacks.
Regulatory navigation: Interpreting ambiguous regulations and making judgment calls on compliance.
Managing AI agents: Ironically, the new "junior analyst" role is overseeing and quality-checking AI output.
How Other Banks Are Responding
JPMorgan's COIN system processes 12,000 contracts in seconds. Morgan Stanley's AI assistant handles 50,000 advisor queries daily. Citadel's AI-driven quant strategies now manage $60B.
This isn't one bank's experiment. It's an industry-wide transformation happening in real time.
How to Position Yourself
If you're entering finance in 2026:
- Learn to prompt and direct AI agents — this is the new Excel proficiency
- Develop judgment and taste — knowing when AI output is wrong matters more than producing it
- Build client skills early — the human layer is the moat
- Understand AI limitations — hallucinated financials in a pitch deck is a career-ending event
The Bigger Picture
Goldman isn't cutting analysts because the work disappeared. The work still exists — AI just does it better and cheaper. The question for every knowledge worker isn't "will AI take my job?" It's "which parts of my job can I hand to AI so I can do the parts AI can't?"
