About this research
This site is a living record of AI-native holding companies: firms that buy or build ordinary businesses and then rebuild how they operate with software and AI, rather than selling software to them. It covers the holding companies themselves, the investors and company builders behind them, the businesses they acquire, adjacent vendors, and the reporting and commentary that surrounds them.
It is not a census, a ranking, or an audit of operating results. It is organised so that a newcomer can learn the category from the ground up and an informed reader can reach the original source for any statement in a small number of steps.
How to read it
Every material number or statement is a claim with an evidence label and one or more sources. Company-reported figures stay labelled as company-reported. A journalist repeating a press release is not a second independent measurement. Forecasts are never treated as actuals. Unknown values stay unknown, and a blank cell in a comparison means the research has not established the value, not that it is zero.
Learning chapters and company assessments are labelled editorial analysis: synthesis written for this site from the reviewed sources. They are not Johnny Perkins’s essays; his bylined writing lives under Essays and is opinion, not evidence.
What is covered
- 8 companies with a full research dossier, and 41 organizations classified as holding companies or platforms overall (many are candidates with limited coverage, marked as such).
- 15 investors and company builders, 7 acquired or operating businesses, and 25 adjacent vendors and comparators.
- 317 reviewed sources and 154 unreviewed leads in the News & Ideas archive, supporting 336 published claims and 28 summarised author views.
These counts are calculated from the categories named above. Publishers and source organizations are not counted as covered companies.
Corrections and revisions
Corrections create revisions; they do not silently erase earlier assertions. Superseded claims remain reachable from the claim that replaced them. Imports are repeatable and keep stable identifiers, so a corrected packet updates the affected sections without republishing everything.
For evidence labels, access statuses and the import rules, see Methodology.