LinkedIn looked at three years of AI hiring data and found something most L&D departments have missed.
AI job postings roughly doubled. The typical AI posting lists $177,000 against $80,000 for non-AI roles. Those numbers will get the headlines. They are not the interesting part.
The interesting part is what companies are actually hiring for. Yes, they still hire people to build AI systems. But the fastest growth is in people hired to deploy AI, manage it, and apply it inside real work. Forward Deployed Engineer is now the third most common AI occupation in LinkedIn’s data. Applied AI roles, which LinkedIn defines as supporting “AI adoption and integration into core company processes,” keep gaining share. So do AI Manager and VP of AI roles, which the report reads as growing demand for oversight as firms scale what they have built.
Read that plainly: the market is paying a premium to close the gap between having AI and getting value from it.
That gap is a workforce capability problem. It is not a content problem, and it is not a tool-training problem. Teaching employees which buttons to press in Copilot was never going to close it.
This is the distinction most organizations still blur: AI literacy versus AI capability.
Literacy means an employee can use the tools. Capability means the employee can point AI at the right problem, judge whether its output is correct, decide where human judgment must stay in the loop, change how the work gets done, and show the performance improved. Literacy is becoming table stakes. Capability is the competitive advantage.
The report gives this distinction economic weight without intending to. It documents two layers of the AI workforce: the people who build the systems, and the people who apply them inside operations. There is a third layer the report does not measure, because it deliberately excludes everything outside builder occupations. That third layer is the organization itself: the managers who set new performance expectations, the workflows that get redesigned, the governance that decides what the technology is allowed to touch, and the employees whose judgment determines whether AI output becomes a good decision or an expensive mistake.
That third layer is where AI investments currently go to stall. Every failed or stalled deployment I see has the same shape. The technology works. The organization cannot absorb it. Nobody decided who owns the new workflow. Nobody defined what good output looks like. Nobody verified anything. Then leadership concludes AI was oversold, when the actual failure was never building the human capability around the tool.
So the workforce question has changed. It is no longer “who knows how to use AI.” It is “who can create business value with it.” That is a harder question, and it is the right one. It forces four decisions most companies have skipped:
First, diagnosis before execution. Generative AI made producing an answer extraordinarily cheap. It did not make solving the wrong problem cheap. Problem diagnosis comes before AI execution, every time.
Second, judgment and verification. If AI produces more of the work, someone has to determine whether the work is correct. “Can employees judge the AI output” is now a core workforce capability, not a nice-to-have.
Third, work allocation. What should humans own? What should AI assist? What should agents execute? Where must a human verify? That is workforce architecture, not training.
Fourth, adoption and evidence. A capability that does not change performance is not a capability. It is a course completion.
None of this is solved by another AI course. The traditional L&D response to every technology shift has been to produce content about it. The labor market is telling us the value moved elsewhere: to the people who integrate AI into work and lead its adoption. L&D has a choice. It can keep manufacturing courses about AI, or it can build the human capability system around AI. One of those has a future.
Companies have spent heavily teaching machines what they can do. The next competitive advantage will come from teaching organizations what humans, AI, and agents should do together.
That is the work.
Going further
The method behind this series is the EDGE framework, Extract, Design, Generate, Evaluate. My book AI for Learning & Development works the full system with the prompts and workflow templates: https://www.amazon.ca/dp/B0D5YJ1QQG
Dr. Ravinder Tulsiani is the author of AI for Learning and Development: Faster Content Isn’t Better Learning and Solve the Right Problem: A Practical Guide for L&D. Take the AI-Ready L&D Function Assessment.
