April 4, 2026
Agent swarm or agent scam?
A case study in testing with 100+ Claude agents in parallel
100 robot testers at once: fans curious, critics cry costs and copyright
TLDR: mngr runs 100+ AI agents to turn tutorial steps into real tests and even refine its own docs. Comments erupted over who owns AI‑made code, whether token costs explode at scale, and if this is just a sales pitch; confusion over tmux added memes, while fans cheered “100 interns” energy.
mngr’s new case study shows off a swarm of 100+ Claude “robot interns” turning a simple tutorial script into real tests, each agent spinning up to run, debug, and polish code while the system stitches the results together. The devs admit end‑to‑end testing is hard, so they let agents try, fail, and improve—and even run them in a terminal multiplexer (tmux).
Cue the comments. Legal hawks warned that AI‑written code may be hard to “own,” with petcat asking if open source licenses even matter anymore. That sparked an IP panic vs keep shipping split: some predict secrecy will replace sharing, others argue documentation and tests still matter regardless of who “owns” the lines.
Money folks arrived with calculators. maxbeech argued the real cost isn’t the number of agents—it’s the schedule: hourly runs and overnight batches mean token bills that “change fast,” citing 20–50k tokens per agent on real projects. Meanwhile, dakolli called it a product pitch.
And the memes? Yokohiii’s “what the hell?” at tmux became the catchphrase, while kanjun dropped a dystopian mic about “agent ragebait” and an AI oligopoly. Fans countered with “100 interns writing tests so you don’t have to.” Welcome to the swarm.
Key Points
- •mngr orchestrates 100+ parallel agents to generate, run, debug, and improve tests derived from a tutorial script.
- •Tutorial blocks from tutorial.sh are transformed into one or more pytest functions to cover multiple scenarios.
- •Each test cites its originating tutorial block via a fixture API, and a script validates that every block is referenced.
- •A slash command (sync-tutorial-to-e2e-tests) packages the conversion and validation workflow.
- •The test framework uses Python’s subprocess module to run CLI commands and capture outputs for assertions.