The On-Ramp Is Closed

The Bureau of Labor Statistics projects 17 percent employment growth for software engineers through 2033. Those citing this as evidence nothing structural is happening are not entirely wrong. They are describing the weather while the climate changes. Entry-level postings dropped 67 percent between 2023 and 2024, per Stanford's Digital Economy Lab, using actual payroll data. For developers aged 22 to 25, employment fell nearly 20 percent. Workers aged 30 and over grew 6 to 12 percent. AI isn't eliminating software engineering. It's eliminating the entry point, which is an entirely different problem on a longer timeline. Salesforce hired no new engineers in fiscal year 2026. Google and Meta are hiring 50 percent fewer new graduates than in 2021. When companies adopt AI tools, junior employment drops 9 to 10 percent within six quarters while senior employment barely moves. The same decision, made simultaneously, which is how you get a trend without anyone having to announce one. Junior developers aren't hired to do junior work. They're hired to become senior engineers. The job is the training program. You learn to debug a production incident at 2 AM by being the person on call when that happens. You develop the judgment that separates a senior engineer from a convincing code generator by being wrong in production until you understand why. Microsoft calls this "AI drag": the finding that AI tools hinder developers who lack the judgment to evaluate what they generate. The downstream consequence is knowledge debt: shallow understanding that holds in demos and collapses in production. McKinsey forecasts a 14 million senior developer shortage by 2030. The industry optimized away the process by which senior engineers are made and would like to register surprise at this outcome. CS enrollment dropped 8.1 percent in 2025, the steepest decline of any major in six years. Now fewer people are choosing to enter the field at all, which is a perfectly rational response and a catastrophic one. If progress stagnates the tools stop improving but the missing cohorts don't come back. If progress continues the bar keeps rising without a clear path to clear it. The situation is alarming in both directions, which is uncomfortable and, it must be said, technically impressive. Engineers who understand why an architecture survives its second year, who evaluate AI output rather than accepting it, who carry judgment from real production systems, are more differentiated than ever. AI has made the lower end of the skill distribution easier to imitate. The upper end has never been harder to reach. The on-ramp is closed. That deserves to be named as such rather than papered over with reassurances about how technology always creates new jobs eventually, which is true in the same way the tide always comes back in and about as useful to someone currently underwater. The road isn't closed. It's narrower. And it no longer starts where it used to.