Espen Skogen has spent two decades inside major financial institutions watching a pattern repeat: when something breaks in technology delivery, the cause traces back to people, judgment, and structure. The technology itself is rarely the problem.
Now he sees a new version of that pattern forming, and this one plays out on a ten-year delay.
The industry is removing entry-level engineering roles at scale because AI handles the work those roles used to do. The logic looks sound on a quarterly budget. The junior developer who wrote boilerplate, fixed small bugs, and built minor features is now outperformed on cost by a model subscription.
Eliminate the training ground and you eliminate the next generation of seniors.
Skogen's view, formed across trading platforms, risk systems, and execution infrastructure built for tier-one institutions, is that this is a structural miscalculation with a long fuse. Senior engineers are made through years of contact with real systems, real failures, and real consequences. Remove the years of contact and the seniors stop appearing.
There's a widely held assumption that senior talent is an input you can purchase when needed. Skogen's model of the industry treats it as an output of a long production process. Every senior engineer he has worked with accumulated their judgment the same way: by doing thousands of small, unglamorous tasks under supervision, getting corrected, and slowly internalizing why certain decisions age well and others accrue technical debt.
That apprenticeship happened inside entry-level roles.
The junior wrote the migration script and learned why schemas matter. The junior broke the build and learned why testing matters. The junior sat in the incident review and learned why architecture matters. None of this appears on a job description. All of it produces the person you eventually trust with a trading platform.
This matters more in finance than almost anywhere else. Research shows that 73% of fintech startupsfail within three years over preventable regulatory issues, while teams with cross-domain expertise on board secured funding 2.8 times faster. Judgment across domains compounds. It cannot be downloaded.
Skogen himself is a product of exactly this process. A qualified solicitor with an information systems background who spent years as a hands-on developer and architect before running RocketFin. The expertise stack took decades to build, layer by layer, through direct contact with the work.
Remove the layers and the stack never forms.
The consequences of cutting junior roles today arrive in roughly five to ten years, which is precisely why the decision feels free right now. Costs drop this quarter. The shortfall lands on someone else's tenure.
Skogen frames this as a familiar failure mode in enterprise decision-making: optimizing a visible metric while quietly destroying an invisible one. The visible metric is headcount cost. The invisible one is the pipeline of people who will be capable of owning critical systems in 2032.
The evidence on what happens when organizations misjudge people-and-structure questions is already extensive. Over $547 billion of AI investment failed to deliver intended business value by the end of 2025, with financial services posting an 82.1% failure rate, the highest of any industry. And 84% of those failures were leadership-driven. The technology worked. The organizational judgment around it did not.
💡 The pattern worth noticing: AI failure in finance is overwhelmingly a judgment failure. The industry's response is to reduce the number of people being trained to develop judgment.
Skogen is careful here in a way that reflects how he analyzes systems. The shortfall is possibly a temporary blip. Markets correct. Institutions eventually notice scarcity and respond to it. The honest position, and the one he holds, is that nobody has a clean answer to what replaces the learning-by-doing model in the interim.
That uncertainty deserves more attention than it's getting.
If the old apprenticeship is disappearing, something has to substitute for it, and Skogen's read is that the replacement looks considerably more academic than the industry expects.
The reasoning runs in a chain. AI will make the software work, in the sense that it compiles, passes the demo, and ships. Working software and maintainable software are two separate properties, and AI optimizes reliably for the first one. Code that works today and becomes a maintenance nightmare in eighteen months is now the cheapest thing in the world to produce.
The scarce skill becomes recognizing that trajectory before it plays out.
That skill is built on things the industry has historically treated as theory:
The next generation learns these formally, through structured study of patterns and their trade-offs, because the informal route of absorbing them across a decade of junior work is closing. Skogen views this as a genuine shift in what technical education is for. The curriculum stops teaching people to produce code and starts teaching people to evaluate it.
The craft moves up a level. It shifts from writing the code to knowing whether the code is worth keeping.
RocketFin's entire operating thesis is that small teams of senior specialists outperform large groups of generalists, and the research base supporting that thesis keeps expanding. Analysis of software project data concludes that 3-7 people is the optimum team size, with teams of nine or more measurably less productive. Competence concentrates. Coordination overhead compounds against you.
This model depends on a steady supply of genuinely senior people. Which makes the pipeline question existential for anyone building the way Skogen builds.
His conclusion cuts against the panic. The AI transition strengthens the case for the senior-heavy model rather than weakening it. When code generation becomes nearly free, evaluation becomes the entire value proposition. The engineer who can look at ten thousand generated lines and identify the three architectural decisions that will cause a production incident under market stress is worth more in this environment than in the previous one.
You cannot review what you never learned to build.That single sentence contains the whole problem, and it's why the training question refuses to go away even as the tooling improves.
⚠️ The trap for institutions is treating evaluation skill as something that emerges naturally. It emerges from deliberate structure: reviewed work, real accountability, and exposure to consequences. Any organization that wants seniors in 2032 has to fund the making of them now, in whatever form the new apprenticeship takes.
Skogen's editorial position, stated plainly, is this: the industry is making a rational short-term decision with a predictable long-term cost, and the institutions that break from the herd will capture the returns.
The break looks specific. Keep a small number of structured development roles alive, redesigned around evaluation rather than production. Pair emerging engineers with senior specialists on real delivery work. Treat design pattern literacy and architectural judgment as first-class hiring criteria for early-career people, on par with what raw coding output used to be.
This is an investment with a long runway and compounding returns. The organizations that make it will hold the scarce asset when the shortfall arrives. The ones that skip it will bid against each other for a talent pool they collectively chose to shrink.
The industry has been here before with judgment failures. The 88% of transformations that miss their original ambitions share a root cause with this one: decisions that look sound at the point of approval and hollow at the point of consequence.
The question every technology executive in finance should sit with is uncomfortable and simple. When the systems your institution depends on need someone capable of judging whether the code underneath them is worth keeping, where exactly will that person have learned how?