Rogo raised its latest round at a valuation that pushed the company past $2 billion. The pitch, repeated across every piece of coverage, was simple: banks run leaner. Fewer juniors needed. The model-building, comp-pulling, deck-formatting grunt work that used to require a stack of analysts gets automated away. It’s sitting inside daily workflows now at more than 250 institutions, including Rothschild, Jefferies, Lazard, Moelis, and Nomura.
That’s the story on the fundraising deck. The story on the desk is different.
What was supposed to happen. Fewer incoming analysts. Smaller junior classes. Associates and VPs doing more with a thinner bench underneath them, because the AI absorbs the volume work. This is the framing that’s shown up across coverage of AI in banking for two years now — Goldman Sachs research has floated the idea that AI could automate roughly a quarter of current banking work hours, with modeling and deck formatting among the most exposed tasks. Reporting via Fortune on McKinsey’s QuantumBlack has described banks cutting incoming junior analyst classes by as much as two-thirds. On paper, the math works: less junior headcount, more AI, same output.
What’s actually happening. Talk to people using Rogo day to day and a messier picture shows up. It’s mostly MDs and Directors who’ve adopted it — plugging in prompts to speed up first drafts of comps, market updates, deal summaries. And it’s associates who are spending real hours quietly fixing what comes back out: a wrong EV pulled from a linked source, an off date buried in a footnote, slide language that reads just a little too smooth and generic to survive a client review without a rewrite. The tool marketed to shrink the junior class is currently generating junior-level cleanup work — it just moved a rung up the ladder, landing on associates instead of analysts.
That tracks with what candidates are also sensing about entry-level hiring right now. Wall Street Oasis threads on this year’s return-offer cycle describe some banks making unusually deep cuts to summer-to-full-time conversion — driven by headcount allocation, not intern performance — while other shops and groups are still converting close to 100%. Conversion rates that typically sit in the 70-90% range are landing noticeably lower at some firms this cycle, and the explanation candidates are hearing isn’t “the intern class was weak.” It’s “we don’t have the seats.” That’s consistent with a bank telling its junior pipeline the AI story while quietly needing more, not fewer, seasoned associates and VPs to actually run the output through a competent human filter.
Why this makes sense once you think about what Rogo actually is. It’s a large language model wired into a bank’s internal data — still a large language model, with the same failure mode every other one has: confident, fluent, plausible-sounding output that’s sometimes just wrong. An analyst who pulls the wrong EV or misreads a footnote gets caught by a VP who’s seen that exact mistake a hundred times and knows where to look. Rogo makes the same category of mistake, at volume, and somebody senior enough to catch it still has to be the one reading the output line by line. The independent benchmarking backs this up — a 2026 test of AI agents on real banking workflows found even the best-performing model failed nearly half its evaluation criteria, and human bankers rated none of the output as client-ready without further work. That’s not a knock on Rogo specifically; it’s the current ceiling for what any AI tool can do unsupervised in this kind of work. The gap between “drafts something quickly” and “produces something a client can see” is exactly the gap a trained associate or VP has to close.
So the actual shift isn’t fewer people. It’s a different shape of demand. Fewer junior seats, maybe — but rising, not falling, demand for the associates and VPs who know the model well enough to catch it when it’s wrong. That’s a much less flattering story than “AI replaces junior bankers,” and it’s a much more expensive one for the banks that bought the first version of the pitch. You don’t save headcount cost by cutting analysts if the analysts’ work just gets reassigned upward to people who cost three or four times as much per hour.
From where I sit doing the hiring, this shows up as a very specific signal. Every search I’m running right now for banks integrating these tools skews toward seasoned associates and VPs — not more junior seats to “manage the AI,” but people senior enough to catch it when the AI is wrong. Clients aren’t asking me to find people who can operate Rogo. They’re asking me to find people who’ve been in the seat long enough to know instantly that an EV multiple is off, that a footnote date doesn’t match the source document, that a synergy number in a draft doesn’t tie out — the same instinct a good VP already has reviewing an analyst’s work, just redirected at a machine instead of a person. That’s a genuine, growing category of demand, and it’s a different hire than the “we need more junior bodies” recruiting that used to dominate this business.
So — is it actually better than an analyst? Depends what you’re asking it to do. On raw speed and volume, yes, easily — Rogo will pull twenty comps or draft a market update in the time it takes an analyst to open the right terminal windows. On judgment, no, not close. It doesn’t know which assumption is aggressive for this specific sector in this specific market, it doesn’t know the client relationship history that makes one framing tone-deaf and another one land well, and it makes the same category of factual slip an overtired second-year makes — except at higher volume and with more confidence in the delivery. A first-year analyst is slower but trainable; six months in, they start catching their own mistakes. Rogo doesn’t get better at your specific deals over time the way a person does — it’s the same tool with the same blind spots on day one and day two hundred. The honest comparison isn’t “AI versus analyst.” It’s “AI as a very fast first-year who never learns, still needs a VP checking every output” versus “an analyst who’s slower now but who you’re actually developing into someone who eventually needs less oversight, not more.” Banks that bought the first framing are discovering they still need the second thing — they just need more of it at a more senior, more expensive level than they planned for.
What this means if you’re staffing a group or thinking about your own career path. If you’re a VP or senior associate right now, “I can catch what the AI got wrong before it goes to a client” is becoming a real, differentiated skill — not a given. If you’re advising junior talent, the honest read is that the tools aren’t shrinking the need for people who deeply understand the mechanics of a model; they’re shifting where that understanding has to sit, from the person building the first draft to the person who has to sign off on it. And if you’re a bank that sold your own board on lower headcount costs from AI adoption, it’s worth checking whether the hours actually disappeared, or just moved to a more expensive person’s calendar.
#InvestmentBanking #AI #Rogo #WallStreet #Recruiting #JuniorBankers #TalentStrategy