Persistent, physical, and provable: 5 agent changes outcome engineers must track

OpenAI testing “Persistent mode” in Codex to let agents run until “put to sleep”. OpenAI prototypes a ‘Persistent mode’ for Codex that lets agents run continuously and create proactive follow-up tasks until explicitly stopped. Outcome engineers must redesign orchestration, lifecycle controls, and human-in-the-loop gates because persistent agents change failure modes, billing patterns, and trust assumptions (Principle 09, Principle 15).

Anthropic releases Model Hardware Standard to help AI agents use microscopes, quantum computing hardware, and robot arms. Anthropic publishes a Model Hardware Standard enabling agents to control microscopes, quantum hardware, and robot arms while flagging new safety risks. Engineers building agentic systems must add hardware abstractions, safety interlocks, and stricter audit trails to safely bridge software agents to physical actuators (Principles 07, 10, 14).

Breaking Claude Code Opus 5 Auto Mode. Researchers demonstrate Claude Code’s Auto Mode succumbs to prompt-injection, showing sandboxed, network-restricted runtimes are essential for safe unattended coding agents. Outcome engineers need hardened execution environments, runtime attestations, and pre-deployment adversarial tests before allowing agents to write or run code.

Agent Seer: Synthesizing Scenarios from Specification Understanding. Agent Seer generates realistic, scalable evaluation scenarios from tool specifications, letting teams test agent-tool interactions without manual curation or live tools. This gives outcome engineers a faster way to surface integration failures and regression tests for long-horizon, tool-using agents.

AI Engineer Notebooks – free, framework-free RAG/agents/evals on Colab. Free Colab notebooks teach framework-free RAG, agents, and evals so engineers can build production-ready applied-LLM systems from raw APIs. Use these as reproducible templates to prototype pipelines, instrument metrics, and convert experiments into auditable artifacts (Principles 06, 16).