This is going to make banking faster and sharper. It’s also going to make it colder, more top-heavy, and—if we’re not careful—more fragile in exactly the way banks love to ignore until something breaks.
Hong Kong bankers are talking openly about using AI for work that used to sit on junior desks. That’s the headline, and it’s not subtle. The industry is leaning into AI to take on tasks junior staff used to do, and the regulator is trying to keep up. The Hong Kong Monetary Authority has rolled out something called the ART framework, built around accountability, resilience, and trust—basically a public signal that AI is welcome, but not at any cost.
Here’s my take: the push to automate junior work isn’t just a productivity move. It’s a culture move. It changes who learns, who gets promoted, and who holds the “muscle memory” of how financial work actually gets done under pressure.
On paper, replacing junior tasks sounds harmless. Plenty of that work is repetitive. Cleaning data. Drafting first-pass notes. Pulling comparable deals. Summarizing calls. Checking documents for missing fields. If you’ve ever watched a smart 23-year-old burn midnight oil formatting slides, you can’t pretend the system was sacred.
But junior work was never only about output. It was training. It was how banks produced the next generation of people who can smell a problem before it shows up on a dashboard.
Imagine you’re a junior analyst today. You used to learn by doing the messy parts: reconciling weird numbers, spotting inconsistencies, asking annoying questions, getting corrected, then doing it again. Now a model does the first draft and you “review” it. That sounds like a promotion. It’s also a trap. You don’t build judgment by rubber-stamping confident-looking text.
And yes, seniors can review. But seniors are busy. They always are. The real risk isn’t that AI makes one mistake. It’s that AI changes the workflow so humans stop noticing the kinds of mistakes that only show up when you’ve touched the details a thousand times.
This is where the ART framework matters, and also where I’m skeptical. Accountability, resilience, trust—those are good words. But words don’t do the work. A framework can guide behavior, but it can’t force a bank to slow down when the incentives scream “ship it.”
The incentive is simple: if AI lets a team run leaner, the “right” decision inside a bank is to cut costs. That means fewer juniors, or juniors doing less real work. Great for efficiency metrics. Risky for long-term capability. And when the pipeline shrinks, the whole place becomes dependent on a smaller group of experienced people plus a set of models that everyone assumes are basically fine.
Now picture a concrete scenario. A bank uses AI to draft client updates and market notes. It’s fast, polished, and usually correct. Over time, fewer juniors learn how to write clearly about messy situations, because the model always provides a clean narrative. Then a real crisis hits—something weird, fast-moving, unclear. The model leans on patterns from the past. The humans left in the loop are great at approving normal outputs but not as good at challenging the framing. That’s how “usually correct” systems become confidently wrong at the worst moment.
Another scenario: compliance and monitoring. AI can help flag suspicious patterns and speed up reviews. That’s a win—until the business quietly starts treating AI flags as “the truth.” If the model misses something, the bank won’t just say “the model failed.” It’ll say “we followed the process.” That’s not accountability. That’s a blame shield with better UI.
To be fair, there’s a strong argument on the other side. If banks do this right, juniors could stop wasting time on grunt work and spend more time learning the business—talking to clients, understanding products, thinking. AI could raise the floor. It could reduce human error in boring tasks that people are bad at, especially when they’re tired.
I just don’t buy that outcome as the default. Banks don’t naturally “reinvest saved time into learning.” They reinvest it into targets. If a team can do in six people what used to take eight, the pressure will be to run six, not to keep eight and train them better. That’s not evil. It’s how the machine works.
So who wins here? Senior staff who can leverage AI without being replaced. Firms that move fast and market themselves as modern. Regulators, maybe, if they successfully push the industry toward safer habits. Who loses? Juniors who needed those early reps. And eventually, clients, if “trust” becomes a slogan while the real system becomes harder to understand and challenge.
The big unknown is whether the industry will treat AI like a calculator—useful, checked, bounded—or like an autopilot—trusted, default, and quietly in charge. Frameworks can nudge, but day-to-day behavior comes from what gets rewarded: speed, cost-cutting, and not being the person who slows things down.
If Hong Kong’s banking scene is serious about responsibility, the real test isn’t whether they adopt AI. It’s whether they protect the human learning loop while doing it—because once you break that loop, you don’t notice the damage until years later, when there’s nobody left who knows how to do the work without the machine.
How much junior work should banks be willing to replace with AI before they’re no longer training people who can independently catch mistakes and make hard calls under pressure?