The Productivity You Gain, and the Generation You Leave Behind

Somewhere in the last two years, a quiet consensus formed in the boardrooms of technology companies. The consensus goes roughly like this: AI-assisted development tools are real, the productivity gains are measurable, and the rational response is to do more with fewer people — beginning at the bottom of the org chart.

The logic is not wrong. The problem is that it is incomplete. And the part it omits will matter considerably more to most organizations in five years than the efficiency they are capturing today.

I say this as someone who has led engineering teams for more than two decades, across organizations ranging from early-stage startups to established R&D operations. I have watched companies make staffing decisions that looked obviously correct in the near term and quietly catastrophic in the medium term. The decision to eliminate junior developer hiring in the age of AI has the same structure. It is extractive thinking dressed up as strategic thinking, and the two are easy to confuse when the quarterly numbers look good.

What the Tools Actually Do

Before addressing the leadership question, the technology deserves an honest accounting, because the productivity gains are real and the skeptics are behind.

AI-assisted coding is not a shortcut for the careless. It functions more like pair programming with an exceptionally capable collaborator — one that demands, above all else, clear thinking. Vague instructions produce vague results. Precise thinking produces output that can genuinely surprise you. A controlled lab experiment by researchers from Microsoft Research, GitHub, and MIT's Sloan School of Management found that developers using GitHub Copilot completed a structured programming task 55.8 percent faster than those working without it.¹ Independent research on enterprise adoption has replicated results in roughly the same range. These are not marginal gains. They represent a meaningful restructuring of what a single experienced engineer can accomplish in a day.

There are legitimate caveats. Code quality does not improve automatically with speed. An analysis of 153 million changed lines of code found that code churn — lines rewritten or reverted within two weeks — was on pace to double in AI-assisted projects relative to the pre-AI baseline, raising real concerns about long-term maintainability.² The discipline required to capture productivity gains responsibly is not trivial. Organizations that have deployed these tools without governance frameworks are already discovering this.

But the productivity signal, properly understood and properly managed, is genuine. Senior engineers using AI assistants are more powerful than they have ever been. The case for adoption is not in question.

The case for unreflective adoption — for treating AI productivity gains primarily as a license to stop hiring junior talent — is where the analysis breaks down.

The Business Case That Gets Ignored

The numbers on junior hiring are not ambiguous. Entry-level hiring at the fifteen largest technology firms fell 25 percent between 2023 and 2024.³ Postings for entry-level positions across the broader U.S. labor market declined roughly 35 percent from early 2023 levels, with AI-exposed roles showing the sharpest drops.⁴ A Harvard working paper tracking 62 million workers across 285,000 firms found that after companies adopt generative AI, junior employment falls approximately 7.7 percent relative to non-adopting firms within six quarters — while senior employment continues to grow.⁵ The mechanism is not mass layoffs. It is a quiet freeze: companies are simply not opening junior positions. The door is closing through inaction rather than decision.

The short-term arithmetic is straightforward, and I have no interest in pretending otherwise. If a senior engineer with AI tools can produce what previously required a team, the case for expanding junior headcount looks weak on a quarterly model. Finance will tell you so. The board will agree.

What the quarterly model does not capture is the talent pipeline. Every experienced engineer — every senior developer, every tech lead, every engineering manager — was once a junior developer learning to think about software by doing the unglamorous work: debugging other people's code, writing tests, maintaining legacy systems, understanding why decisions made three years ago now cause problems. These experiences are not supplementary to professional development. They are the mechanism by which professional development occurs. You do not become a senior engineer by reading about what junior engineers do. You become one by doing it, badly at first, under conditions that allow for recovery and growth.

The entry-level tasks being automated first are, in a precise sense, the same tasks that have served as the profession's training ground for decades. When those tasks disappear from the junior role, the role loses its developmental function. And when the role loses its developmental function, companies stop hiring for it. The result, playing out right now across the industry, is a severing of the bottom rung of the career ladder — not just for this generation of graduates, but for the talent supply that organizations will need in 2030 and beyond.

This is the calculation that is missing from most boardroom conversations about AI and workforce. Companies are optimizing for today's headcount against today's capability requirements. They are not modeling the compounding effect of a multi-year interruption in junior talent development on their future senior talent supply. In a sector where the half-life of technical skills is already short, this is not a minor oversight.

The Pipeline Is Not Someone Else's Problem

There is a version of this argument that treats the talent pipeline as an industry-wide commons — a problem that will be solved collectively by the market, by universities adapting their programs, by new types of roles emerging. That version is comforting and probably partially correct over a long enough horizon. It is not, however, a strategy for the individual organization.

Consider what happens at the firm level over a five-year cycle. An organization that cuts junior hiring significantly in 2024 and 2025 will have meaningfully fewer internal candidates for mid-level roles in 2027 and 2028. It can attempt to hire experienced mid-level engineers from the market — but those engineers are increasingly scarce, because other organizations made the same decision. The result is a competition for a shrinking pool of experienced talent, at escalating cost, to fill roles that could have been developed internally at a fraction of the price. The short-term efficiency gain funds the medium-term talent shortage.

I have observed a version of this pattern before AI entered the picture — in companies that eliminated internal training programs to reduce costs, then paid a premium to hire externally for capabilities they had previously developed in-house. The logic was identical: the cost of training felt discretionary; the cost of external hiring felt unavoidable. In both cases, the decision made sense on the spreadsheet and did not make sense in practice.

History offers a precedent worth examining. When Henry Ford more than doubled his workers' wages to five dollars a day in 1914, his primary stated reason was unsentimental: a roughly 370-percent annual turnover rate was destroying the productivity of his assembly line, and higher pay was, as Ford later described it, one of the finest cost-cutting moves the company ever made.⁶ The workforce investment was not altruism. It was a recognition that the short-term expense of developing and retaining people was substantially cheaper than the long-term cost of chronic attrition. The principle has not changed. What has changed is the form in which companies are now incurring it.

What Leadership Actually Requires Here

Acknowledging this problem does not require abandoning AI adoption. The tools are too valuable and the competitive pressure too real for that to be a serious option. What it requires is making the investment in junior talent a deliberate line item rather than a casualty of efficiency.

In practical terms, this means several things. It means maintaining some threshold of junior hiring even when senior productivity gains make the immediate business case for it look weak — treating it as an investment in future capability rather than a current-period cost. It means redesigning the junior role itself, recognizing that AI has changed which tasks are available for it: the work is not gone, but it has shifted toward review, evaluation, prompt engineering, and quality judgment — skills that still require experience to develop but develop differently than they did before. And it means building explicit mentorship and development structures into AI-augmented team workflows, since the informal knowledge transfer that used to happen organically in teams doing grunt work together does not happen automatically in teams where most of the grunt work is delegated to a model.

None of this is easy to maintain when every efficiency metric is pointing in the other direction. That is precisely why it requires deliberate leadership rather than default optimization. The organizations that will have the most capable senior engineering teams in 2030 are the ones whose leaders today are willing to hold two things simultaneously: the genuine power of AI-assisted development, and the genuine obligation to the people whose careers depend on being given a first chance to learn.

AI gives experienced engineers more leverage than they have ever had. Whether that leverage is used to quietly close the door on the next generation, or to develop them faster and more effectively than any previous generation, is not a technology decision. It is a leadership decision.

And right now, most organizations are making it without realizing they already have.

SK102 applies AI-assisted development to build software for Guam organizations — with local engineers who train the teams that inherit the systems. Learn how we work.