Mizuho trains bankers to code; HSBC plans job cuts

Mizuho Financial Group will launch the "Mizuho AI Bootcamp" in November, a three-month program aimed at transforming sales and administrative staff into system developers, according to a report from Nikkei that has been picked up by Asian media outlets this week. The initial class will consist of about 20 people. The stated goal is to requalify several thousand professionals over the next ten years.

Eight days earlier, the Financial Times reported that HSBC plans to cut around 70% of financial consultants and half of the management and specialist positions in its wealth management business in the UK, with layoffs expected by the end of October. The bank did not confirm the figures but noted it continues to adapt its UK operations to more digital products and channels.

It is tempting to view Mizuho as the virtuous case and HSBC as the predatory one. Both perspectives overlook what matters most to those who sell services to banks.

The Japanese Pilot and Its Limits

The Mizuho program is born from a small pilot. Between June and August, five employees participated in the pilot, four of whom had no prior programming experience. All met the required standard and were transferred to engineering roles in September, according to reports. The bank argues that generative AI tools shorten the learning curve for those who have never written code, and that hiring AI talent in the Japanese market has become challenging.

Five individuals do not validate a model. The website Economy.ac, in its analysis of the plan, points out the risk of an overly optimistic outlook: if the new developers can't identify errors and vulnerabilities in the AI-generated code, the burden of review falls back on experienced engineers, which is precisely the scarce resource the bank seeks to conserve.

There is also a fact that weakens the reading of Mizuho as a bank that "does not cut jobs." In February, according to Nikkei Asia, the group announced its intention to eliminate work equating to up to 5,000 administrative positions in Japan through AI over ten years. The retraining of several thousand coexists with the reduction of thousands more. The difference lies in the pace and mechanism, not the outcome.

The American Equation

Bank of America adds a variable that neither of the two cases captures. During the earnings call on October 14, CEO Brian Moynihan stated that the bank hired about 1,000 contractors to fix software flaws identified by advanced AI models, including Mythos from Anthropic. They work alongside approximately 3,000 employees in that area. Moynihan estimated the cost of remediation and anti-money laundering compliance for the quarter at "several hundred million dollars, upward," and expressed hope that these expenses would decrease over time.

In the same call, he projected that the roughly 130 internal AI projects at the bank will save $700 million by 2027, up from $300 million this year. In the American case, AI creates a new human workload before it eliminates another. Those who only read the economic side miss the transition cost; those who focus solely on the cost overlook the profit curve the CEO is prepared to defend before analysts.

The Axis Ignored by the Debate

The question "does AI cut or create jobs in banks?" is framed incorrectly. The three cases demonstrate that it does both within the same balance sheet. The variable that distinguishes Mizuho, HSBC, and Bank of America is who absorbs the transition cost and over what time frame.

In Japan, the long-term employment norm and the scarcity of engineers drive the bank to internalize this cost, retraining staff and spreading reductions over a decade. In the UK, a more flexible labor market allows HSBC to make adjustments in weeks, with layoffs scheduled for the end of the month. In the United States, Bank of America outsources the peak human demand that its own AI has generated.

Stuart Breyer, in an article in the British magazine Money Marketing, argues that HSBC's cuts should not be seen as a model for the financial advisory profession. It is the right objection to the enthusiasm of those already predicting the end of human advisors. The symmetrical objection applies to those who view the Japanese bootcamp as proof that AI preserves jobs: Mizuho itself maintains its reduction target.

For consultancies and integrators, each model generates a different demand. The Japanese approach invests in training, code review, and quality governance for years. The British model buys digital service platforms and quick fixes. The American model purchases remediation hours at scale, with tight deadlines. A single supplier catering to all three needs three distinct commercial proposals, and the most costly mistake is to offer a bank a transition it cannot absorb.