This is AI Frontiers. Today, who actually pays programmers and why? Firms do not pay for coding skill. They pay for proximity to revenue, control, and liability. That is the core argument. AI does not change the mechanism. It makes it visible, and it widens the gap between those who define systems and those who execute inside them. Start with the most cited productivity number. The Peng et al. controlled experiment on GitHub. Copilot found that developers completed one bounded JavaScript task 55.8% faster than a control group of 95 professional developers. That was a lab style task, not production telemetry. Strategic depth and code quality did not rise at the same rate. Routine implementation gets faster. Judgment does not automatically compound with it. A 2025 randomized trial by Meter found something sharper. Experienced open source developers working in their own mature codebases were actually slower with early 2025 AI tools, while believing they had been faster. The U.S. Bureau of Labor Statistics projects software developer employment to grow 15% from 2024 to 2034, with roughly $129,200 openings per year. Median annual pay for software developers was $130 in May 2024, large numbers. But they flattened the difference between software as a core competitive asset and software as administrative overhead. The useful taxonomy is not front-end versus back end, it is proximity to economic control. System architects define boundaries and shape future option value. Product engineers translate commercial ambiguity into coherent software under real constraints. Implementation specialists execute bounded tasks inside established frameworks. That last layer is easiest to benchmark, template, and compress. AI accelerates that compression. Generative tools are strongest on bounded work, where patterns are common and success can be checked quickly. They are weakest where the task is choosing trade-offs under uncertainty and seeing second-order effects across systems. For employers, faster output creates durable value only if someone still controls architecture, testing standards, security assumptions, and failure handling. If that control layer is thin, AI raises the speed at which organizations accumulate technical debt and compliance risk. The B plus trap applies directly. AI lowers the cost of producing plausible looking code. It does not lower the cost of being wrong in production. One strong technical lead can now direct more implementation work than before, whether done by junior staff, contractors, or AI-assisted teams. That increases the span of control at the top and weakens the position of workers whose contribution sits mainly in straightforward execution. The most important shift may not happen in the editor at all. It happens in procurement terms, liability clauses, security schedules, and model usage rules. Contracts now determine more software behavior than many engineers want to admit. That's AI frontiers.