Which Roles Seed-Stage AI Companies Posted

Methodology and data

Who is included

Companies building AI-native products that announced a seed or pre-seed round in the last 12 months, 5 to 150 employees, selling to businesses or developers. Not consultancies, not agencies, not open-source projects without a company.

This is a tracked sample, not a verified complete census. 124 companies are included; 37 have observed posting history and 105 have a headcount reading. A company without a found posting may still be hiring through other channels. Company shares use all 124 tracked companies as the denominator.

Collection limit: the 20 most recent postings per company; 17 companies reached that limit. Counts describe observed postings, not unique vacancies or completed hires. Reposts can represent the same search, and limited histories cannot establish a complete company ranking or a hiring growth rate.

124 companies in the census; 105 with a headcount reading; job boards read for 0 (the live board read has not run for this panel yet, so open roles are unknown, not zero); 489 postings over the last twelve months for 37 of them, the 20 most recent a company (17 companies at that cap, so earlier quarters are undercounted); signals cover the last 180 days. Fact sheet computed 2026-10-17; study written 2026-10-17; dataset version vertical@1.0.0; intelligence rules intelligence@0.2.1; claim check: pass. The executive summary, the sections, the FAQ answers and the limitations were revised by Human in the Loop Talent on 2026-10-20 against this fact sheet: Re-checked against the 2026-10-17 fact sheet. The generated headings and FAQ questions were renamed, so every override is re-keyed. Postings moved from 490 to 489. The six-month funding count rose from 8 to 18, but the new company-event matches include news about other organisations with similar names, so the revision no longer quotes a cohort funding count and names only the two pre-seed rounds the sheet attributes to the company itself. Layer's seed round is no longer in the sheet and is dropped. The generated "Who these companies are" section asserted that these companies have no recruiting function and that most postings are first hires; neither is in the data, so it is replaced. The earlier corrections stand: the archive is not described as current, AI engineering's recent share is read on its small base, and unmapped leadership seats are not read as findings.

How the census and the article were made

Original research computed from the signal layer of the recruiting CRM behind Human in the Loop Talent: public job boards, company news, funding announcements and headcount readings, collected continuously for every tracked company and recomputed every Sunday. A figure is published only when the sample supports it; the data behind every page is published as JSON next to it.

1. The category is defined in one sentence (who is in, who is out). A search plan turns it into a structured company-database query (keywords, products, funding stages, founding year, countries, size band) and six to ten web-search angles.
2. Three feeds enumerate the companies: the structured query against a company database; the list pages a researcher would read for each angle, fetched whole and read for every company they name; and bulk firmographic enrichment (headcount, industry, headquarters) for any company that arrived without them. The lists merge by company domain, then by name.
3. Companies that pass the inclusion gate (category, size band, geography) become tracked companies in the CRM, so the signal layer watches the whole category: job postings every weekday, news, funding and leadership changes. The last twelve months of each company's postings are read from a job-postings archive, so the study sees a year of hiring, not only what is open today.
4. After at least a week of observation a fact sheet is computed in code: headcount bands, twelve months of hiring by role family and quarter, open roles, open leadership searches, tools named in job posts, revenue-seat changes, funding and AI initiatives over 180 days.
5. The study is written from the fact sheet only. A second pass checks every number, name, title and date against the sheet and strips anything it does not contain; the claim-check result is shown on the page.
6. A census under ten companies is not published. Studies are rewritten when the census is at least a week old and the last study is older than four weeks.

Key figures

The family whose postings lean most into the last 90 days, measured below the cap: AI / ML engineer (55% of its 11 uncapped postings, at 4 companies; a flat year would put 25% there).

Postings by quarter are not shown: with 17 companies at the archive cap, earlier quarters are undercounted and a quarterly series would read as growth that is an artefact of the cap.

Limitations

This study covers the 124 companies in the seed-stage AI census, 105 of which have a headcount reading. The study reads a twelve-month archive of job postings, not live job boards, so it reports what was posted and when rather than which roles are open today. The hiring figures come from a twelve-month archive of postings to 2026-10-17 covering 37 companies and 489 postings, capped at the 20 most recent per company; 17 companies sit at that cap, so their counts are floors and no quarter-by-quarter trend can be read. The sixty-day figures are posting dates from that archive, not a read of what is open. Leadership seat holders are not mapped, so nothing is known about CRO changes. Company events (funding, AI initiatives) cover the 180 days to 2026-10-17 and only the sources the CRM reads; they are matched to companies by name, and some matches are about other organisations with similar names, so this page quotes only events that name the company itself. Months named in the text are the dates the postings appeared in this archive.

• 87 of 124 companies have no postings in the twelve-month archive. That is unknown, not idle: the archive covers public job boards and does not reach every company.
• The archive keeps the 20 most recent postings per company and 17 of 37 companies with postings sit at that cap. Their counts are floors, their older quarters are undercounted, and a quarter-by-quarter trend cannot be read from this panel.
• Open roles have not been read live from these companies' job boards, so no row states a current open-role count. "Posted in the last 60 days" (220 roles at 36 companies) is a posting date in the archive or the daily board reads, not a live read of the board.
• Company events cover the 180 days to 2026-10-17 and only the sources the CRM reads (company news, LinkedIn posts, ColdIQ); an event missing here may still have happened.
• 105 of 124 companies have a headcount reading; the rest are counted in the census without a size.
• Nothing on this page states that a hire caused growth or that a company will hire; the panel records correlations and the page says which ones.

Dataset and versions

• Fact sheet and study: /research/verticals/seed-stage-ai-companies.json (version vertical@1.0.0, computed 2026-10-17).
• Company opportunity table: rendered on this page by rules intelligence@0.2.1; not offered as a separate download until redistribution terms for the underlying posting archive are confirmed.
• Revision log: 11 revisions since 2026-10-10, read from the git history of the published JSON.
• Licence: CC BY 4.0. Reuse with attribution.

Cite this

Related: all intelligence, which roles precede growth, startup recruiting, post-funding hiring.