Why 90% of Claude Code Users Are Missing Its Most Powerful Feature
Source: ZIRU, cited in Medium Daily Digest of 2026-04-07 (Email 23). Direct URL: medium.com/@jh.baek.sd/why-90-of-claude-code-users-are-missing-its-most-powerful-feature-52ef453785ed (12 min read, 50 claps)
Key takeaways
- Framing: the secret isn’t better prompts or smarter models — it’s the harness/workflow layer most users ignore.
- Lines up with the wider “Agent Harness” vocabulary Joe Njenga and Rick Hightower also use.
- Useful for convincing skeptical engineers that the harness is the product.
- Pair with Agent Harness and Superpowers / Cialdini for the full story.
The argument
ZIRU’s piece sits alongside Njenga’s “Agent Harness” article as part of the April 2026 convergence on what the community is finally calling out: you cannot prompt-engineer your way past a poorly-configured harness. CLAUDE.md, hooks, skills, MCP servers — the layer between you and the model is where 90% of adoption is stuck.
Why the framing works internally
When advocating for harness investment inside your team, this article is useful rhetorical support — “90% are missing it” is a compelling frame for allocating engineering time to CLAUDE.md hardening and skill library curation.