Research

Analyses

Each of these is generated from the live panel and carries its own method and limits. They are registered in the database, so this list is what exists rather than what someone remembered to link.

• What AI-related postings ask for that others do not
  Within one discipline, what changes about a job posting when it mentions AI?
• Which occupations does each discipline prepare students for?
  Given a business discipline, which SOC occupations does it lead to, how strongly, and on what evidence?
• What do those occupations pay, and how big are they?
  For the occupations a discipline prepares students for, what are the national median wages and how many people hold them?
• Does the finding replicate outside the panel built around AI employers?
  Run the same comparison separately in four sub-corpora assembled for different, and in one case opposite, reasons. Which term groups move in all of them?
• How much AI do entry-level postings ask for?
  Does the AI-related share of postings differ by seniority band, and does it differ the same way in a panel built around AI employers and one that was not?
• What share of each discipline's postings mention AI? withheld
  Across the twelve business disciplines, what proportion of postings are AI-related?
• The four quadrants, program by program
  Within each business program, how are postings split across the four AI-capability quadrants, and what does each quadrant ask for?
• Where do graduates of this programme actually go?
  Which employers hire from the programme, how often, and how consistently across years?
• How much of where graduates go can this panel actually see? withheld
  Of the employers that hire from the programme, what share appear anywhere in the job-postings panel?
• Where is AI NOT showing up in my programme, and what is asked there instead? withheld
  For each MSB programme, which of its job categories are AI-thin, and what do those postings ask for in place of AI?
• What did each programme's graduates actually apply for, and actually accept? withheld
  For each MSB programme and graduating class, which job titles and employers did students apply to, and which did they accept?

A panel of public job postings, collected to answer one question: what does AI-related business work actually ask for? Everything below is computed from the current classifier run when you load the page. Figures that have not cleared their evidentiary bar are named and withheld rather than quietly omitted.

The current snapshot

Run 4c4ad6ed, postings first seen 2023-03-01 to 2026-11-11. Boards are collected hourly and these aggregates rebuilt on the same schedule, last at 2026-11-11 14:18 UTC.

Four sub-corpora, never pooled

These are four different collections with four different shapes, and adding them together produces a number that means nothing. The ATS panel is a live sweep of applicant-tracking systems whose board list was assembled because those employers post AI work, so its AI share is a property of the sample. The Muse and the remote boards were not selected for AI at all, which is what makes the gap between them worth looking at. The Hacker News “Who is Hiring” archive is the only part reaching back before 2026, and its title field is just the first 140 characters of a comment, which makes title-based classification unreliable there.

What is withheld, and why

Classifier v6 checked against 232 reference codings made by a language model, not yet by people, so the figures below measure agreement with another machine rather than accuracy. Macro F1 0.79. Percentages are withheld below n = 200; rows below n = 20 do not appear at all.

Each count is the postings that involve the discipline. A posting can involve several disciplines and none is ranked first, so it is counted once in each, and the rows add up to more than the corpus.

What AI-related postings ask for that others do not

Within a single discipline, on a single job board, comparing postings that mention AI against those that do not. Both groups pass through the same classifier, so a misclassification lands a posting in the wrong discipline but keeps it on the correct side of the AI split. That is why this table is published while the absolute per-discipline shares above are not.

A lift of 2 means an AI-related posting is about twice as likely to ask for something as a comparable posting that does not mention AI. 1 means no difference. Below 1 means AI-related postings ask for it less. That sentence is the whole of it, and it was missing from this page for as long as the page has existed.

Lift is adjusted for posting length, so it is not the ratio of the two percentages beside it. AI-related postings run longer than other postings, and a longer posting trips more of any term list for reasons that have nothing to do with the job. Each lift below compares postings of similar length and pools those comparisons, which moves every figure down. Dividing the two percentages instead gives the unadjusted number this site published before 11 September 2026.

Only mappings signed off against the AI Engagement Model appear here. An element whose term set has not been ratified shows nothing rather than a provisional number.

All business disciplines

n = 12,232 AI-mentioning and 2,841 other postings. Percentages are the share of postings carrying any term in the group; counts in brackets are the postings behind each percentage. The lift column adjusts for posting length and is therefore smaller than those two percentages divided.

Does it replicate?

The same comparison run separately in each sub-corpus. These panels were assembled for different reasons, and The Muse was not assembled around AI at all, so it is the closest thing here to a control. A term group that moves in every column is describing AI-related work. One that moves only in the first column is describing how the panel was built.

A dash means the cell did not clear the suppression rules in that sub-corpus, usually because fewer than 5 postings carried the term in one arm. Columns are never summed: they are different populations, not parts of one. Lift is length-adjusted, as above.

By discipline

A ratio needs cells, not just a base. A group appears only where the discipline has at least 50 non-AI postings and at least 5 postings carry the term on each side — the rule that would have stopped an earlier draft reporting a 3.31 lift built on one posting.

Greenhouse boards only, so the compared groups share a platform, a collection window and an employer pool. Restricted to each posting’s primary discipline.

Counting rules that change the totals

A posting counts for every discipline that can legitimately claim it. A product manager counts under Business Management, Information Systems, Marketing and Strategic Management at once. Discipline totals therefore exceed the number of postings, by design.

Never sum across disciplines and call the result a job count. When a partition is needed, only the primary relation is used, and the multi-label overlap is reported beside it rather than folded in.

What this panel is good for

The sharpest check on the last point: none of the nine employers the anchoring school’s placement records name as top recruiters for Finance and Strategic Management appear anywhere in this corpus. The firms that actually hire those graduates largely do not post on these platforms.

Sources

Every posting is collected from a documented public API. Sources are grouped by why they are in the panel, because that governs how their numbers may be read.

The theory this feeds

The reason to ask what employers want is to know what to teach. The AI Engagement Model sets out the eight modes, three tiers and the dispositions underneath them, with the evidence walked through step by step.

Read the theory walkthrough on aimodes.ai · How this panel is built · Decisions behind it