Resilience

At 2:44 p.m. UTC on March 11, 2026, thousands of developers found themselves locked out of Claude Code mid-session. No warning. No graceful degradation. Just a dead CLI and a 15-second timeout loop. The good news: if you were connecting via API key, you noticed nothing. The Claude API stayed fully operational throughout the two-hour incident — a detail that matters enormously for anyone designing resilient agentic workflows. What Actually Broke The failure was isolated to OAuth authentication — the browser-based login flow that Claude Code uses to connect to Anthropic’s servers. When developers ran /login, their browser would open, they’d click “Authorize,” see a confirmation… and then the CLI would hang until hitting its hardcoded 15-second timeout. ...

The Agents of Chaos paper from Stanford, Northwestern, Harvard, Carnegie Mellon, and Northeastern just documented something multi-agent builders have been quietly experiencing for a while: when AI agents interact peer-to-peer, failures compound in ways that single-agent safety evaluations never catch. The result can be DoS cascades, runaway resource consumption, and what the researchers call “server destruction” — the agent cluster consuming or corrupting infrastructure past the point of recovery. This guide covers the practical patterns that prevent that outcome. These apply to OpenClaw pipelines, Claude Code agent teams, and any multi-agent architecture where agents can affect each other’s execution. ...