Skillselion Research · Edition of 2026-08-12
The Agent Use-Case Census
What are AI agents used for? We skipped the surveys and weighed 164,841,042 installs across 79,848 listings and 54 categories. The single largest use is making images and video, and the most crowded category earns the fewest installs per listing.
On 2026-08-12 we dumped the whole Skillselion production catalog and summed installs per category. Coding agents earn their name: ten software-engineering categories together take 30.1% of installs. The single hottest thing people install, though, is media generation, and the category builders flood hardest pays worst. You can reproduce every number from the dataset; the method and its limits are in the methodology.
Finding 01
The install-weighted answer: media first, then planning, then code
Rank all 54 categories by installs and Generative Media leads with 34,347,034 installs, 20.8% of the entire ecosystem, from only 1,469 listings. Productivity & Planning follows at 10.4%. The first classic coding category, Backend & APIs, sits in eighth place at 4.1%. People point their coding agents at pictures, videos and plans more than at any single kind of code.
- Generative Media (1,469 listings)20.8%
- Productivity & Planning (3,410 listings)10.4%
- AI & Agent Building (18,087 listings)8.1%
- Automation & Workflows (2,805 listings)7.6%
- Cloud & Infrastructure (1,417 listings)6.1%
- Design & UI/UX (2,049 listings)5.0%
- Office & Documents (719 listings)4.3%
- Backend & APIs (4,553 listings)4.1%
- Testing & QA (2,281 listings)3.5%
- Marketing & SEO (2,014 listings)3.4%
Finding 02
Code work is the biggest cluster. Media is the biggest single use.
Group the categories and the picture sharpens. Ten software-engineering categories (Backend & APIs, Frontend Development, Testing & QA, Code Review & Quality, DevOps & CI/CD, Databases, CLI & Terminal, Documentation, Security, Cloud & Infrastructure) sum to 49,548,598 installs, 30.1%. Generative Media alone takes 20.8%. Office work (Productivity & Planning plus Office & Documents) takes 14.7%. So yes, agents work on software. And still, one category of one kind of output outweighs any single kind of engineering.
- Software engineering (10 categories)30.1%
- Generative Media (one category)20.8%
- Office & productivity (2 categories)14.7%
The buckets are editorial groupings of catalog categories, and they are listed in full above so you can rebuild or redraw them from the dataset.
Finding 03
Builders flood the meta category. Installs go elsewhere.
AI & Agent Building is the most crowded category in the catalog: 18,087 listings, 22.7% of everything published. It collects 8.1% of installs. One listing in four is tooling for building more agents, while the audience spends its installs on media, plans and code. Supply concentrates where builders live; demand concentrates where their users do.
Finding 04
Agents mostly help people build, not launch or grow
Every listing sits on one stage of the builder journey, from idea to operate. The build stage takes 64.0% of all installs (105,532,154across 53,431 listings). Shipping takes 12.1%, operations 7.5%, growth 6.5%, and launch just 2.5%. The agent economy today is a construction economy. The launch-and-grow half of the journey barely registers, which reads like a map of what agent builders have not automated yet.
- Build (53,431 listings)64.0%
- Ship (8,901 listings)12.1%
- Operate (5,502 listings)7.5%
- Grow (4,325 listings)6.5%
- Validate (2,298 listings)4.1%
- Idea (3,831 listings)3.4%
- Launch (1,560 listings)2.5%
Method
Methodology
The census is a complete dump of the Skillselion production catalog taken on 2026-08-12: 79,848 listings and 164,841,042 cumulative installs, aggregating public registry data, chiefly the skills.sh registry, alongside GitHub and other public sources. Each listing carries exactly one of 54 catalog categories and one journey stage, assigned during ingestion and enrichment from the listing’s own documentation; both columns cover 100% of rows. Installs count registry install events, not active users, so the distribution measures developer-side adoption through agent tooling registries rather than enterprise seats.
Two groupings on this page (software engineering, office) are editorial sums over named categories, and the names are printed next to each number. Single-category assignment cuts both ways: a video skill used for marketing counts once, under its primary category. The dataset carries the per-listing rows, so any alternative grouping is one pivot table away.
Skillselion is an independent project and is not affiliated with Anthropic, OpenAI, Cursor, Claude, Claude Code, or Codex.
Data
Download the dataset
The full census runs 79,848 rows, one per listing, with type, category, journey stage and installs. It is free to use with attribution under CC BY 4.0: cite “Skillselion Agent Use-Case Census, August 2026” and link this page.
FAQ
Questions journalists ask
What are AI agents used for?
Measured by installs rather than surveys: making images and video leads at 20.8% of the 164,841,042 installs in this census (category: Generative Media). Classic software-engineering work (ten categories from Backend & APIs to DevOps & CI/CD) takes 30.1% combined, office work (productivity, planning, documents) takes 14.7%, and building more agent tooling takes 8.1%. The full 54-category distribution is on this page and in the dataset.
How is this different from an AI agent use-case survey?
Surveys record what people say; this census records what they installed. Each of the 79,848 listings carries one catalog category, and we sum cumulative registry installs per category. The bias to know about: installs measure developer-side adoption through agent tooling registries, not enterprise seat counts.
What counts as one use case here?
The catalog category of a listing, one per listing, 54 categories total. Categories are assigned during ingestion and enrichment from each skill’s documentation. Two editorial groupings on this page (the software-engineering bucket and the office bucket) name their exact member categories, so you can recompute or regroup them from the dataset.
Can I reuse these numbers or the dataset?
Yes. Cite "Skillselion Agent Use-Case Census, August 2026" and link this page. The raw dataset (79,848 rows with category, journey stage and installs per listing) is downloadable on this page under CC BY 4.0.