At 95% per-step reliability, a 30-step agent job finishes clean only 21% of the time
Scobleizer · x · 2026-09-08
Scobleizer highlights VC18z's sober counterargument to the "unstoppable agent" demos:
- Reliability math: at 95% per-step reliability, a 30-step job only completes cleanly about 21% of the time — error compounding is the core weakness of long agent chains.
- Verification cost: verification can wipe out the speed win agents promise.
- Write access: giving agents write permissions turns an assistant into infrastructure with a much larger attack surface.
- Positioning: Salesforce, GitHub, and Jira already own the workflow — is the agent a new software layer, or just a new UI on the old one?
The author calls this the key underwriting question for agentic software over the next 24 months.
More from coding & agent
- AI code review blasted for flagging Carmack's inverse sqrt in 10 lines of working code — ZeeshanZiaML · 2026-09-08
- WSL Manager 2.0 ships a built-in MCP server letting agents create and run Linux distros — bostrot · 2026-09-08
- DHH's Omarchy Linux launches foundation with $15.5M, betting on agent-powered troubleshooting — vista8 · 2026-09-08
- CROCODIL tackles over-editing when LLMs modify code written by other models — jessyjli · 2026-09-08
- How to Auto-Generate Figma Photomosaics with Grok Bot, Cutting 90% of the Grunt Work — mattyp · 2026-09-08
- x402 gives agents price tags per call, enabling economic reasoning, says Coinbase dev lead — kleffew94 · 2026-09-08