43%. That's the share of CEOs now planning to cut junior roles because of AI, up from 17% just one year ago, according to an Oliver Wyman survey. The same survey found only 27% of those CEOs said their AI investments were producing measurable results.
That 16-point gap - cutting people at nearly double the rate of proven returns - is the central tension in enterprise AI right now.
Spending Without a Clear Return
Uber is the clearest case study. The company burned through its entire 2026 AI budget by April, four months into the year. At that point, 95% of their engineers were using AI tools and 70% of code commits were AI-generated. Their COO still couldn't draw a clear line between that volume of usage and actually shipping more useful features to users.
This isn't a uniquely Uber problem. Microsoft and Duolingo have both pulled back from aggressive AI-driven workforce plans in recent months. Neither cited poor tool performance. Both pointed to the difficulty of translating usage metrics into outcomes that justify the investment.
The underlying issue is structural. The productivity gains from AI tools are real but diffuse - spread across dozens of small time savings, faster code reviews, marginally better first drafts. Those gains don't aggregate into a number you can point to in a board meeting. The costs, by contrast, are concentrated and legible. That asymmetry makes the business case genuinely hard to make even when the tools are working.
The Junior Developer as a Convenient Answer
Junior developers are the easiest cut in this environment. They're a fixed salary attached to a defined skill set that AI tools now partially replicate. Cutting them is fast, visible, and makes sense on a budget sheet in a way that "we got 15% more efficient across the board" doesn't.
The problem is that junior developers provide things AI tools don't. They accumulate institutional knowledge about why the system works the way it does. They catch errors in AI-generated code, which matters more when 70% of your commits are AI-written. And they're the pipeline: your senior engineers five years from now come from this cohort.
When you compress that human review layer while expanding AI usage, you're not just cutting costs. You're reducing oversight precisely when you've introduced more AI-generated code that needs reviewing. The failures that result tend to be subtle - wrong behavior rather than obvious crashes - and expensive to diagnose.
The companies that survive this period intact are likely the ones that found a way to measure what AI actually changed about their output, not just their usage. The ones making workforce decisions based on the assumption that 70% AI commits means proportionally fewer junior developers needed are accumulating a deferred cost that's difficult to see until it's very expensive to fix.