If AI is economically valuable, why doesn’t that consistently translate into demand for engineers who can build it?

Assume the following:

• AI systems are generating real economic value (cost reduction, productivity gains, new products)
• There exists a subset of engineers who can genuinely build and extend these systems (model training, fine-tuning, agent design, tool integration, etc.)

Given those assumptions, why does the labor market not reliably reflect strong demand for that group?

Specifically:

• Why isn’t the marginal value of those engineers clearly reflected in hiring outcomes?
• Is the value being captured primarily by capital (infrastructure, model ownership) rather than labor?
• Are hiring mechanisms (credential filtering, risk aversion, signaling) obscuring real capability?
• Is there a structural lag between where value is created and where firms are willing to hire?

I’m not questioning whether AI itself is valuable. I’m trying to understand why that value does not appear to propagate cleanly to the people capable of building it.