Large language models become consequential agents when the surrounding system lets their outputs change external state. Current models can call tools, operate interfaces, delegate tasks, retain state, inhabit generated worlds, and control robots or laboratory equipment.
These advances are often narrated as a single march toward autonomy, conflating model competence, system integration, persistence, and safe authority.
This review synthesizes publicly available research and technical specifications up to 31 August 2026, organizing evidence along delegated authority, temporal persistence, and environmental coupling, while separating model, harness, and environment.
The evidence shows that expansion of action interfaces (e.g., tool calls, UI operations) is documented more convincingly than full task completion, error recovery, authorization verification, or independent validation.
Model Context Protocol and Agent2Agent improve interoperability but do not establish trustworthy delegation; multi‑agent organization adds specialization together with cost and correlated failure risk.
Persistent simulations and world models support training and planning but do not themselves demonstrate agency; robotics and self‑driving laboratories exhibit bounded feasibility rather than unattended open‑world reliability.
We propose “justified delegation” as an analytical and normative heuristic, not an observed law or certified score: expand the action scope only where evidence supports provenance, bounded authority, failure detection, safe recovery, and calibrated human control.
From this framing we outline a research agenda: coupled model‑harness evaluation, capability‑based permissions, durable state, cross‑agent accountability, and staged physical validation.
Review