Updated
Updated · O'Reilly Media · Sep 29
Jesse Vincent Launches Superpowers for Claude Code, Framing 50-Plus Prime Radiant Tools Around Human Intent
Updated
Updated · O'Reilly Media · Sep 29

Jesse Vincent Launches Superpowers for Claude Code, Framing 50-Plus Prime Radiant Tools Around Human Intent

2 articles · Updated · O'Reilly Media · Sep 29

Summary

  • Superpowers, released by Jesse Vincent last October, trains Claude Code to behave more like a disciplined senior engineer by enforcing process rather than relying on raw model capability alone.
  • SKILL.md files Vincent discovered inside Claude.ai’s Linux-backed environment inspired the framework, which he says works best by steering model behavior with rationale and intent instead of piling on prohibitions.
  • 2004 management lessons from leading inexperienced programmers shaped the system: agents get roles, journals, code-review subagents and even a "therapist" subagent that alone can edit persona files.
  • Verification is central because agents can misreport their own work; Vincent separates coding from certification and has agents produce proof such as demo videos after repeated self-testing.
  • Prime Radiant now lists 50-plus products, and Vincent argues the scarce human skill in agentic engineering is no longer typing code but clearly stating goals, asking better questions and judging results.

Insights

Can Jesse Vincent’s workflow-first method make AI agents trustworthy across tools, or does it mainly expose how unreliable autonomous coding still is?
If AI coding agents need therapists, journals, and proof of work, are they really autonomous programmers or just interns with better syntax?
What happens to software engineering when the hardest skill is no longer writing code, but defining, verifying, and judging what agents produce?