The tools AI job posts ask for, by role

Every month a script reads the public job boards of the same 100 employers that build or sell AI: model labs, AI-native apps, AI infrastructure companies, and established tech, data and fintech companies. It reads Ashby, Greenhouse and Lever boards through their public APIs and keeps only each posting's ID, title, link, date and the terms it names.

Each posting goes to one of eight roles based on its title. The roles are checked in a fixed order: forward deployed engineer, GTM engineer, solutions engineer, AI engineer, RevOps, SDR or BDR, account executive and customer success. Managers, directors, heads, interns and recruiters are left out. Customer success managers, account managers and account directors stay in, because at AI companies these are individual jobs, not people managers.

Engineering roles count at every employer on the list. Sales, RevOps and customer success roles count only at the AI companies, so the sales picture is about selling AI. If one employer posts the same title and description in several locations, it counts once.

The dictionary has 277 terms of three kinds. Tools are named products you can learn and put on a resume, such as Salesforce, Clay or Claude Code. Foundations are what most postings in a field assume, such as Python, SQL, AWS or Excel. Practices are the work itself, such as building agents, running evals or a sales methodology. Names that can mean more than one thing count only in context: Conductor only next to coding agents, Segment only as the data product and Go only as a programming language.

A term counts once per posting. The script reads each posting's headings and notes where the term appears: in the requirements, in the responsibilities, in a nice-to-have list or somewhere else. A sentence that calls a skill a plus, a bonus or preferred counts as nice to have wherever it sits. When a posting names a term in more than one place, the strongest place wins: required first, then responsibilities, then nice to have. A term in the job title counts as required.

An employer's own products never count for that employer. Names in a list of customers, investors or partners don't count either. Neither does any line an employer repeats in at least 50% of its postings, which is where the about-us, benefits and legal text sits.

The big percentage is the share of employers hiring for a role that name the term in at least one posting. An employer that posts the same job fifty times still counts once. A tool needs at least two employers to be ranked. Share is the fraction of a role's postings that name the term, shown when you open a row. Capped share counts at most the 20 most recent postings from each employer, so one company hiring in bulk can't set the order. Lift compares a role with all eight roles together. A lift of 3 means the role's postings name the term three times as often as the average posting. A signature tool has a lift of at least 2, a share of at least 6% and at least 3 employers.

Emerging marks an AI-native product that people work in or build agents with, released or widely adopted since 2023: coding agents, AI work assistants, AI-native sales and automation tools, and the newest agent SDKs. The radar lists emerging tools that at least two employers name for a role, however small the share.

A role with 40 or more postings is ranked. A role with 20 to 39 is shown with a small-sample note. A role with fewer is not shown.