The Agent Economy Census

Finding 01

Developers installed AI that interrogates them 2.4M times

One whole category exists to give the developer a hard time. grill-me pressure-tests your plan until every decision survives questioning, and it holds 787,155 installs on its own. Across 88 skills built for interrogation, the cluster reaches 2,425,925 installs, and it is the platform’s #1 rising search at +160% over the census window. Nothing else in the census grew faster.

1. 1grill-me (mattpocock)787,155
2. 2grill-with-docs (mattpocock)669,499
3. 3triage (mattpocock)534,688
4. 4grilling (mattpocock)380,693
5. 5doubt-driven-development (addyosmani)14,704

Finding 02

One solo educator rivals Microsoft. OpenAI is 70x behind.

Rank publishers by installs and the order stops matching company size. Microsoft leads with 11,153,893. Second place belongs to one person: TypeScript educator Matt Pocock, at 10,584,924. OpenAI sits at 150,665, roughly 70 times behind that single independent educator. Reach in this ecosystem tracks what developers choose to install, which is a different contest from model benchmarks or funding rounds.

1. 1Microsoft11,153,893
2. 2Matt Pocock (solo educator)10,584,924
3. 3Vercel Labs5,576,605
4. 4Anthropic3,484,550
5. 5Google730,896
6. 6AWS171,446
7. 7OpenAI150,665

Finding 03

A 158M-install economy where 0.07% is for sale

Of 79,827 listings, 55 carry a price. That is 0.07%. Another 1,404 ask you to bring your own API key and 184 run freemium; the rest cost nothing at all. Adoption on this scale normally arrives with a paid tier, a store cut or a marketplace fee, and almost none of that has been built here yet. Attention concentrates regardless: the top 1% of skills take 77.4% of all installs, and 27.2% of skills have fewer than ten.

1. 1No pricing listed (free)74,755
2. 2Explicitly free3,429
3. 3Bring-your-own API key1,404
4. 4Freemium184
5. 5Paid55

Finding 04

The office rituals are already automated

Skills that write standup updates, summarize meetings, triage inboxes, tailor resumes and draft performance reviews reach 2,101,631 installs between them. Standups, meeting notes and status reports account for 1,889,988 of that. Every cluster here automates the reporting of work rather than the work itself.

1. 1Standups, meeting notes & status reports (111 skills)1,889,988
2. 2Email triage & cold-email drafting (38 skills)117,318
3. 3Resumes & cover letters (129 skills)86,914
4. 4Performance reviews & self-assessments (10 skills)7,411

Finding 05

The biggest force in the ecosystem is ByteDance

The largest publisher footprint in this American AI ecosystem belongs to ByteDance. Its Lark/Feishu office suite ships 70 listings carrying 24,261,658 installs, 15.4% of every install in the census, ahead of Microsoft, Anthropic and OpenAI. One caveat we owe you: Lark skills often install as bundles, where a single action installs several skills, so their per-skill counts are not comparable to a hand-picked install. The share is real. The mechanism producing it differs from the rest of the table, and the methodology sets out how.

1. 1ByteDance Lark/Feishu (70 listings)24,261,658
2. 2Microsoft11,153,893
3. 3Matt Pocock (solo educator)10,584,924
4. 4Vercel Labs5,576,605
5. 5Anthropic3,484,550

Finding +

Bonus: inequality beyond anything a country reports

We ran the Gini coefficient, the statistic economists use for income inequality, across the install counts of all 58,751 skills. It comes out at 0.96. South Africa, which reports the highest national income Gini the World Bank publishes, sits near 0.63. Half of all 158M installs go to 198 skills, 0.34% of the catalog. The median skill has 56 installs, 59.3% have fewer than a hundred, and 9.9% have none at all. For most people who publish a skill, release produces no measurable adoption.

1. 1AI agent skill economy (installs)0.96
2. 2South Africa (most unequal country)0.63
3. 3Brazil0.52
4. 4United States0.41
5. 5Norway0.27

Country figures are World Bank income Ginis, latest available. Income and software installs are different measurement domains, so treat the comparison as a sense of scale rather than a like-for-like ranking.

Method

Methodology

The census is a complete dump of the Skillselion production catalog taken on 2026-08-09: 79,827 listings, of which 58,751 are agent skills and the rest are MCP servers, plugins and marketplaces. Install counts are cumulative all-time totals. They aggregate public registry data, chiefly the skills.sh registry, alongside GitHub and other public sources, and they count install events rather than active users.

We built the category clusters (interrogation-style, workplace rituals) by keyword matching over names and descriptions, then curated them by hand to drop false positives. A “brutalist” UI-design skill, for instance, is not an interrogation skill. Publisher totals group by publishing account. Suite publishers such as Lark distribute skills in bundles, so one user action can register many installs, and we flag that wherever it affects a finding.

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 79,827 rows, one per listing, with installs, publisher, category, risk level, quality tier, duplicate flag, license and trend fields. It is free to use with attribution under CC BY 4.0: cite “Skillselion Agent Economy Census, August 2026” and link this page.

Download the census CSV · 79,827 rows · 1.7 MB gzipped