The work exists before the job title does — what 47,101 Fortune 500 postings reveal about AI hiring

The title looks familiar. The work inside it has already moved.

The work arrives first: connect a foundation model to a product, build retrieval, evaluate an LLM, or wire an agent workflow into an existing system. Then comes a harder organizational question: what do you call the person who owns it?

Andela analyzed 47,101 Fortune 500 software postings and scored 2,026 skills against established role boundaries. Among 1,832 postings whose titles named home roles for the relevant bundle, chiefly AI Engineer and ML Engineer, 53% — 972 postings — already carried at least two skills from different established roles.

The title names a role the market already knows. The job description is moving beyond it.

6,758 postings point toward a similar bundle without a settled title

The study detected 23 recurring skill bundles that did not fit neatly inside one established role. Notable examples included MLOps Pipeline Engineer and LLM Application Engineer.

The latter describes an engineer building applications on top of foundation models rather than primarily training the models themselves. Its bundle had coverage across 6,758 postings.

That number is not 6,758 jobs titled “LLM Application Engineer.” Coverage means postings carrying enough skills from the bundle under the study’s test. The useful signal is precisely that similar work is recurring before one title fully owns it.

New jobs do not necessarily begin with new names

A backend engineer picks up LLM orchestration. Data-pipeline responsibilities combine with ML evaluation. The combination repeats across companies, and only later does a market label start to stabilize.

Andela summarizes the phenomenon as job titles lagging the work. Its study is a cross-sectional snapshot of roughly five weeks, however, not a time-series measurement of occupational growth. It detects structural drift from role baselines rather than proving that a named occupation is growing over time.

For employers, “how many AI Engineers?” may be the second question

If model training, LLM application engineering, inference and MLOps, and agent integration all sit under one AI Engineer label, the title may be convenient while obscuring the actual hiring boundary.

A team that needs Software Engineering + RAG + cloud deployment can miss capable Backend or Platform engineers if it searches only for five years of ML Engineer experience. The reverse problem also exists: a broad AI Engineer title can pull candidates who lack the production experience the role actually needs.

Some apparent scarcity can therefore be a search-boundary problem as well as a supply problem. Define the problem and evidence required before assuming the market has no people.

For candidates, searching only your current title can hide your next market

A Software Engineer who has spent two years building LLM APIs, RAG, agent workflows, evaluation and monitoring may overlap heavily with jobs called AI Engineer, Applied AI Engineer or ML Engineer.

When titles are unstable, the résumé needs to explain what was built and operated, not merely how long the candidate carried a particular label. A person may already be doing the work of an emerging role without holding its eventual title.

Indeed shows the opposite motion at the same time: AI is spreading into titles

Indeed Hiring Lab found that the number of US normalized job titles with a meaningful number of postings explicitly mentioning AI in the raw title rose from 264 in 2022 to 822 in Q1 2026. In the US, 63% of those AI-touched titles were outside tech occupations.

So two movements can coexist. New work hides inside familiar titles, while familiar occupations acquire new AI labels. The labor market is not simply producing a list of brand-new occupations; it is rewriting the naming system itself.

In talent search, search the problem before the title

When a role is still forming, searching only for people who already hold its newest title can close the candidate pool too early. The relevant person may still be called Software Engineer, ML Engineer or Platform Engineer.

The search question becomes: who has already solved this company’s problem under another name? Not every AI hiring shortage is a naming failure, and genuinely scarce skills remain scarce. But in a fast-moving market, how the candidate boundary is defined can materially change who becomes visible.