Only 14.6% of 'AI-native' job postings actually name an AI tool. I checked 37,920.

Only 14.6% of 'AI-native' job postings actually name an AI tool. I checked 37,920.

Canonical: this is a cross-post. The original lives at https://four-leaf.ai/blog/what-ai-native-means-in-job-postings

Every company in tech calls itself AI-native, and as a label it's useless until you see what it asks for in writing. So we checked. Four-Leaf's AI Stack Index analyzed 37,920 job postings from public company career feeds between April 1 and early May 2026, deduplicated from 48,053 raw listings and capped so no single employer makes up more than 5 percent of the sample. For each posting we checked whether it names any of 75 AI tools and skills, and whether each is required, preferred, or just mentioned. The full dataset is free under CC BY 4.0, so pull the rows and check any number below.

Only one in seven postings names an AI tool at all

The headline number is the quiet one. Across 37,920 postings, just 14.6 percent mention any of the 75 AI tools and skills we track. The label is on the company. The concrete requirement is on a minority of the roles.

A company can be AI-native in its product and its pitch while most of the jobs it posts ask for the same skills they asked for two years ago.

Even the most common AI skill is small

When AI tooling does show up, it's narrow. The most-mentioned AI skill is agentic AI, meaning agents and agentic workflows, at 8 percent of postings. After that the drop is steep. PyTorch appears in 1.8 percent, retrieval-augmented generation in 1.3 percent, the OpenAI API in 1.3 percent, Cursor in 1.3 percent, prompt engineering in 1.2 percent, and the Anthropic API and TensorFlow in roughly 1.1 percent each.

No single AI tool outside of agentic work clears 2 percent of the market. Chasing a long list of them is wasted effort. Depth in one or two that match your target roles beats a resume that name-drops ten.

And it's almost never actually required

Even where AI tools appear, they're usually a nice-to-have rather than a gate. Agentic AI is mentioned in 8 percent of postings but listed as required in only 0.1 percent. Retrieval-augmented generation, the model APIs, and Cursor are each required in essentially zero percent of listings even where they're named.

AI fluency reads as a tiebreaker, not a barrier to entry. Developers who assume an AI-native company will reject them for not knowing a specific framework are usually wrong about how the postings are written.

The requirement concentrates in a few functions

AI tooling isn't spread evenly across the org. Four functions sit well above the 14.6 percent average. Data roles mention an AI tool 26.7 percent of the time, engineering 26.6 percent, design 24.8 percent, and product 22.4 percent. Marketing is near the average at 16.7 percent, customer and sales roles around 14.5 percent, and it falls off from there, with operations at 7 percent and scientific roles at 5.2 percent.

If you're targeting data, engineering, design, or product, treating one agentic framework and one major model API as table stakes is reasonable. Outside those functions, an AI-native employer is far more likely to care that you use AI tooling in your workflow than that you can name a specific library.

The takeaway

AI-native is mostly positioning until it's read against what the postings require in writing. In 37,920 listings, a named AI-tool requirement shows up in about one in seven roles, the most common single skill reaches only 8 percent, the tools are almost never mandatory, and the demand clusters in data, engineering, design, and product. The full report and dataset back every figure here.