In Where AI Lives we argued that AI lives in four places: built, bought, bundled and brought. This map puts the 2025 federal inventory into those places.
OMB’s 2025 inventory lists 1,040 deployed AI use cases. Of those, 1,013 say how the agency got the system. We sorted them into three groups:
- Built: 556, made by agency staff or with a contractor
- Bought: 342, purchased from a vendor
- Bundled: 115, vendor AI that came inside another product
Brought AI is missing by definition: an inventory only lists what someone reported.
Inside each group, use cases with similar profiles sit together. Built AI spreads across generative AI, language processing, prediction and science, risk and security models, and identity matching. Bought AI leans toward surveillance, investigations and imaging. Bundled AI falls into three pockets with almost nothing in common: office and IT software features, medical devices, and platform and security features. What ties them together is how they arrived, not what they do.
Method
Source labels come from OMB’s contracting_usage field. Built is “In-house Development” or “Contracting and In House”; Bought is “Vendor Purchased.” Bundled is our labeling: vendor-purchased cases where the AI is a feature of a product bought for something else, assigned by rule and reviewed by hand. The 27 cases that left the field blank are not shown.
Each case carries 39 tags drawn from OMB’s own fields and from its description: AI type, topic, impact and PII flags, inputs and tasks. Vendor and product names are not tags. Positions inside each group come from t-SNE on those tags, and the clusters from k-means. The spacing between the three groups is layout only.
Data: Office of Management and Budget, 2025 Federal Agency AI Use Case Inventory, individual-use-case CSV at commit 06c7ebe, filtered to development_stage “Deployed.”