The rush for agentic AI is facing a mismatch in demand for natural language processing (NLP) and "prompt engineering" skills in 2024, according to a new McKinsey measure, but workers familiar with PyTorch and TensorFlow were about equally in demand.

The sudden boom in investment in AI has had analysts scrambling to track demand for specific skills and areas where there are mismatches between supply and demand – and to figure out how much money various skills demand.

PwC's first AI Jobs Barometer in 2024, for instance, found that AI-specialist jobs commanded a premium of up to 25% compared to their closest non-AI equivalents, by looking at job ads across 15 countries.

A skills shortage

Such job-ads analysis is getting increasingly granular, and this week McKinsey offered a new look at mismatches in its Technology Trends Outlook 2025 report. 

Matching the skills pool against job ads, McKinsey found that AI hirers struggled to find people who can wrangle AWS. The skill came in as the third most sought after "machine learning" and Python, but only a tiny number of candidates listed AWS among their capabilities. 

Java skills were nearly as popular as AWS, but Java coders were available to match the demand.

See also: Not the CIO’s job? Getting your organisation AI agent ready – “the promise and the peril”

There seemed to be about half the Python programmers required, and about 90% of the machine learning experts sought.

When it came to agentic AI specifically – in which hiring rocket from a base of effectively zero before 2024, Python ranked as the most important skill needed. Machine learning was some way behind that, and PyTorch and TensorFlow were equally sought-after down the list. 

In agentic AI jobs, the mismatch between demand and available skills was in NLP, which ran short to the tune of 60% – and in "prompt engineering", which McKinsey measured as being available in quantities 3.6 greater than required.

TensorFlow vs PyTorch

TensorFlow, first released by Google in 2015, was in slight over supply compared to the demand McKinsey measured. PyTorch, created under Meta's flag in 2016, was just slightly under-supplied.

PyTorch is often associated with research, while TensorFlow is more closely associated with production systems, though the pace of the growth in AI development in general means it is not clear to what extent each is in use. 

Agentic AI seems set to significantly distort the market, coming from almost nowhere to be a dominant talking point in 2025. 

Last year, McKinsey said, agentic AI attracted $1.1 billion in equity investment, compared to more than $124 billion for AI generally. By some estimates, the agentic AI enterprise market will be worth more than that in revenues well before 2030.

The link has been copied!