30 Days, No Revenue, and the Legal Time Bomb: Zero-Human Companies Hit a Wall

Analysis

A $1M AI Agent Experiment in 30 Days: 238 Endpoints Built, $0 Earned

This is the most honest ZHC case study you'll read — agents built a full API product, listed on MCP registries, published 136 SEO pages, and hit $0 in revenue because bot detection and distribution walls are the real moat, not code. The gap between 'agents can build anything' and 'agents can sell anything' is still enormous, and this repo documents exactly where that gap lives.

14 YC W26 Companies Hit $1M ARR With Sub-5 Headcount — A Repeatable Pattern Emerges

YC W26 is the cleanest data set yet that agent-native isn't a vibe — it's a go-to-market strategy, with 14 companies crossing $1M ARR on 1-4 person teams and 14% weekly revenue growth. The metric that matters here is 'agentic task completion rate,' which investors are now pricing into term sheets.

The Liability Gap: Who Gets Sued When an Agent CEO Approves a Bad Contract?

The IBM refund-agent case and the Air Canada chatbot ruling are early signals that courts are coming for ZHC founders first — not framework maintainers, not model providers. If you're running an agent stack without explicit liability architecture, you're the defendant.

Tools

83,000 Combined GitHub Stars in 90 Days: The Zero-Human Company Repo Cluster Mapped

OSS Insight mapped the entire ZHC GitHub ecosystem — Paperclip, edict, oh-my-claudecode, ClawCompany — and the architectural patterns are diverging fast: safety veto layers, cost-routing by task complexity, and persistent agent memory are the three bets that separate serious frameworks from demos. The Clipmart marketplace and verticalized Paperclip adapters are the next wave to watch.

One Founder, 2,000 New Customers/Month: The Paperclip Marketing Stack Dissected

Postiz founder Nevo David's 6-layer agent marketing stack is the most concrete 'here's what I actually run' account in the space — the key insight is that GitHub Issues as an agent memory and feedback loop is a genuinely clever workaround for the lack of persistent agent context. The honest caveat at the end matters: this isn't zero humans, it's humans as quality controllers instead of executors.