Which process is ready for autonomous AI agents?

Describe a process, answer five more questions — Claude Sonnet 4.6 classifies the right agent pattern, estimates risk and effort, and lists the required safeguards. Honest, hype-free, with a veto rule on critical data.

Important: this evaluation is first-pass orientation, not an architecture spec. Before implementation, a real requirements- and risk analysis with compliance and security owners is required.

Methodology

Six inputs are passed to Claude Sonnet 4.6. The model classifies into one of six agent patterns (single-step / retrieval / tool-use / multi-agent / human-in-loop / not-suitable) and returns reasoning, risk score, automation grade, and required controls. A hard server-side override kicks in: critical processes with sensitive or personal data are never allowed to remain pure multi-agent — minimum pattern is human-in-loop.