What Makes a Role Truly AI-First?
Open any job board and you will find “AI” everywhere. According to Indeed’s October 2026 labor market snapshot, AI-related postings now make up 7.3% of all postings, well past the previous peak of 3.3% in 2022. That is a striking number, and it is worth reading carefully.
What the number counts
Indeed’s methodology note explains that the share is built from keywords: terms such as “Machine Learning,” “Data Science” and “Artificial Intelligence,” plus generative AI terms such as “Large Language Models.” A posting counts if it mentions them.
That is a reasonable way to measure how much AI vocabulary has entered the market. It cannot tell you whether the person hired will spend their days on AI, or whether the word sits in a company blurb, a perks list or a “nice to have” line. A backend engineer who occasionally calls a model API and a researcher training one both show up in the same count.
For a candidate, that gap is the whole problem. You can spend hours on applications for roles that only borrow the label.
The core test
At AIfirst.jobs, an AI-first role is one where AI is the substance of the work. Ask: if you removed the AI or ML component, would the job still exist in a recognizable form?
- AI-first: training and evaluating a recommendation model, building a retrieval pipeline over company data, doing alignment research, running a fine-tuned LLM in production.
- Not AI-first: a general software role at an AI company where the daily work is internal tooling or CRUD APIs, or a role that mentions Copilot in the benefits.
The company is a blunt filter. Most people at an AI company are not doing AI work, and plenty of people doing AI work sit at companies nobody calls “AI companies.” We evaluate each posting on its own: a Claude model reads the full description and applies a strict is_ai_core check. If AI appears only in the perks, the listing is excluded.
Why tasks matter more than titles
Analyst Josh Bersin makes a related point from the employer side. In an October 2026 column on task intelligence, he argues that jobs should be viewed as sets of tasks, each of which needs assessing in the context of the wider role and business outcome. That is his analysis, not a measurement, but the logic carries over to job hunting: read the tasks, not the title.
There is a related signal on skills. In a September 2026 ERE report on an iCIMS survey, iCIMS’s head of talent insights Trent Cotton said most people use AI tools in a basic way, while employers want more specialized skills. The same article reports that using generative AI tools such as ChatGPT or Copilot was the only AI skill reported by over half of workers. Using a tool is common. Building, evaluating and operating AI systems is the narrower thing that AI-first roles ask for.
The 12 categories we use
We classify every role into one of 12 categories:
- LLM Engineer — building applications on top of foundation models
- ML Engineer — training and optimizing ML models
- MLOps — deployment, monitoring, and infrastructure for ML systems
- AI Product — product management and design where AI is the product
- Data Scientist — experimentation and analysis with an AI/ML focus
- Data Engineer — pipelines and feature stores that feed ML systems
- AI Researcher — academic or industry research, publishing, R&D
- Prompt Engineer — prompt design, fine-tuning, RLHF, evaluation
- AI Safety — alignment, interpretability, red-teaming
- AI Infrastructure — GPU clusters, inference hardware, distributed training
- Computer Vision — image/video models, multimodal systems
- AI Generalist — cross-functional roles that don’t fit neatly above
A practical filter
Before you apply, scan the responsibilities for verbs: trains, evaluates, ships, monitors, researches. If AI shows up only in the company description, treat it as a different kind of job. To start from roles that pass the test, browse the feed and filter by the category that matches your skills.