← The AI Hype Audit — all 645 verdicts

PARTLY

Partly true: agency-agents is a real, hugely popular MIT-licensed repo, but its 'agents' are prompt files for your coding tool, not a company that runs itself

Facebook · Nov 2026

The claim'This GitHub repo can help you build an entire AI company, with 232 specialized AI agents organized into 16 different departments.' Over 160,000 stars, with engineering, design, marketing, sales, finance and more departments.

The repo is msitarzewski/agency-agents, and the screen recording in the reel shows it clearly. On Nov 15, 2026 it had about 168,600 stars and an MIT license, so 'over 160,000 stars' is true. The roster numbers are slightly stale: the repo now holds about 283 agent files across 18 division folders (engineering, design, marketing, sales, finance, paid media, game development, security and more); the reel's 232 matches the dashboard in its own footage and was likely accurate when filmed. The oversell is 'build an entire AI company'. Each agent is a Markdown file that gives a persona, rules and deliverables to a tool you already run, such as Claude Code, Cursor, Codex or Gemini CLI. The repo's own installer copies those files into your tool. Nothing in it hires, coordinates, remembers customers, books revenue or runs on its own; you still drive every session, pay for the model, and check the output. Installing all of them at once can also crowd your tool: the README itself warns that OpenCode silently drops agents past about 119. Used well, it is an excellent free library of role prompts. The reel sells nothing directly; it is an engagement reel for a tools page.

What holds up

  • Pulled the repo from the GitHub API on Nov 15, 2026: about 168,600 stars, MIT license, active commits that week.
  • Counted agent files by folder in the repo tree: about 283 across 18 divisions, versus 232 in 16 claimed.
  • Read the README: agents are Markdown persona files installed into Claude Code, Cursor, Codex, Gemini CLI and others by a script or a desktop app.
  • README notes an OpenCode limit of about 119 registered agents and recommends installing by division.

What doesn’t

  • 'Build an entire AI company' overstates a prompt library.
  • Agent and department counts are stale (now about 283 in 18).
  • You still pay for and supervise the underlying model.
  • Installing every agent at once can crowd or break some tools.

The catch

Real repo, real stars, genuinely useful. But it is a library of role prompts, not a company; a person still runs every task.

How to actually do it

  • Clone msitarzewski/agency-agents and run ./scripts/install.sh --list teams to see what is there.
  • Install only the one or two divisions you need, for example --tool claude-code --division marketing.
  • Call one agent on one real task, compare its output to your current process, and keep only the agents that beat it.

The free value is a well-written set of role prompts for the AI coding tool you already use. Treat it as a toolbox, not an employee roster.

Confidence
High
Posted by
an AI tools reel page (about 94,000 plays); reel sent to Buddy Sat Nov 14, 2026, 9:18 PM

See the original claim →

We test hype for free. We build the real thing for a living.

Thirty minutes, no pitch — and you'll leave with something useful either way.

Book a call with Todd or start with the free Business Checkup →

The Verdict Weekly

Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.

© Schreier Group · schreiergroup.com · See a wild AI claim? Drop it here and we'll test it.