Andela research finds AI and ML job postings blend skills, identifies five emerging hybrid titles

Start with what the 53 percent counts. A posting qualifies if it lists at least two skills that Andela's taxonomy attributes to two different established roles ^[3]. That taxonomy spans more than 2,000 skills feeding 23 emerging titles ^[2], an average near 87 skills per title ^[17]. Clearing a two-item threshold in a catalogue that size is not demanding work; most senior requisitions I have written would trip it before the responsibilities section.

The arithmetic narrows further. Fifty-three percent of 1,832 postings is about 971 listings ^[13], and 1,832 out of 47,000 postings is 3.9 percent of the sample Andela analyzed ^[14][1]. So the bundling finding rests on a slice of the corpus, described by a test the article does not define beyond "two skills from different roles" ^[3].

Now look at how the recommended roles are specified. Andela's MLOps pipeline engineer is given as 46 percent ML engineer, 23 percent DevOps engineer, 15 percent data engineer, and 8 percent each AI engineer and data scientist ^[5], which sums to 100 ^[15]. The FinOps reliability engineer is 36 percent DevOps, 27 percent SRE, 18 percent cloud engineer, and 9 percent each DevSecOps and cloud solutions architect ^[7], summing to 99 ^[16]. These are shares of a single skill bundle, not lists of requirements. Every one of the five headline roles is defined as a mixture of three or more established roles, including the LLM application engineer at 48 percent AI engineer and 34 percent ML engineer ^[6] and the product frontend engineer at roughly a third frontend and a third product manager ^[9].

Under the 53 percent test, all five would score as bundled. The distinction Andela is actually drawing is between a bundle that has been named and apportioned and one dumped into a nebulous title ^[4]. That is a real distinction, and the article's own lineage supports it: DevOps came out of friction between developers and operations, DevSecOps out of pushing security through the delivery pipeline ^[20]. Both were bundles too. Naming them is what made them hireable.

For the 53 percent to transfer to your own requisitions, the skill-to-role mapping has to hold, and the article does not publish it ^[19]. If a shared tool such as a container runtime or a query language attributes to both DevOps engineer and data engineer, the test fires on any posting that mentions infrastructure and data in the same paragraph. Without that mapping, the number is not reproducible against your ATS.

The stall itself is not measured. As reported, Andela's research does not track time-to-fill, vacancy duration, or a count of unfilled requisitions ^[18]. Skill overlap in a job description shows how the description was written; whether anyone was hired from it is a separate question.

The claim I would test is Cory Hymel's, Andela's head of research: historically 30 to 40 percent of a DevOps engineer's bundle was role-specific, the remainder was cross-role habitable, and it is the habitable remainder that AI can absorb first, which raises the weight of the specific part ^[11]. Hymel also says the data does not support the vendor line that AI pushes engineers toward generality ^[10]. That is a falsifiable mechanism, and it is checkable on one team in a quarter. Andela is a talent and services platform ^[21], so read the specialization conclusion with the placement business in view.

If you are deciding whether to split an AI engineer requisition, this gives you the vocabulary for the split. The fill-rate case still has to come from your own hiring data.