An ex-CEO of an investment bank just explained what bankers actually do all day, and it's the best preview we have of which white-collar jobs AI will actually kill.
The Summary
- Scott Bok, former Greenhill CEO, breaks down investment banking work in the context of AI disruption
- Before we can know if AI replaces bankers, we need to understand what the job actually is (not the Hollywood version)
- The real question: which parts of elite knowledge work are performance theater, and which parts are irreplaceable judgment
The Signal
Investment banking has always been a black box to outsiders. You know they make deals happen and get paid obscenely well, but the actual day-to-day work? Opaque. Bok's timing here matters because we're past the "AI will change everything" phase and into the "okay, but specifically what" phase.
The interesting part isn't whether AI can build a pitch deck faster (it obviously can) or run comparables analysis (boring, yes). The interesting part is whether the core of investment banking, advising CEOs on billion-dollar bets during high-stakes moments, is actually analyzable work or relationship-based performance art. If it's the former, AI agents eat it. If it's the latter, humans keep the fees.
Here's what nobody wants to say out loud: a huge amount of white-collar work that pays well exists because it signals trust and seriousness, not because it requires genius-level analysis. Hiring Goldman to advise on your merger signals to shareholders and boards that you did your homework. The 200-page deck might be mostly justification for a decision the CEO already made. That's not a criticism, it's how trust works in systems where billions are at stake and someone needs to be accountable.
The agent economy will ruthlessly expose this dynamic. AI can generate the analysis, build the models, stress-test the scenarios. What it can't do (yet) is sit across from a nervous CEO at 11pm and say "I've done this 50 times, and here's what happens next" with enough conviction that they believe you. That judgment, that pattern recognition from lived experience, that's the moat. But it's a narrower moat than bankers want to admit.
The Implication
If you're in a knowledge work job right now, ask yourself: am I paid for analysis or for judgment? For the work product or for the trust signal? The analysis jobs are getting automated fast. The judgment jobs are safe for now, but the barrier to entry for "judgment" will drop as AI gets better at pattern matching across millions of scenarios no human has seen. The real skill in the agent economy won't be doing the analysis. It'll be knowing which analysis to trust, and being accountable when you're wrong. That's a different job description than most white-collar workers signed up for.
Source: Bloomberg Tech