Reliable Execution (Value)

Description

Third motivating value. The agent does what the human actually meant, stays coherent over time, and supports verification before declaring success. Spans single-turn correctness and long-horizon dependability (context-window boundaries, session resume, multi-agent delegation). Grounded in Anthropic's gather-context / take-action / verify-results loop and the harness-design admonition that agents 'confidently praise mediocre work,' motivating separation of generation from evaluation. Motivates context-as-scarce-resource, append-only state, graceful recovery, isolated subagents, and defense in depth.

Key claims

Relations

Sources

src-20260423-0cff68d3291b