Codex is now available inside the ChatGPT mobile app so long-running coding-agent work can be reviewed, approved, and kicked off away from the desk.

The Codex safety writeup emphasizes control surfaces, sandboxing, agent-native telemetry, and audit trails — directly relevant to safe RPA/agent execution.

OpenAI announced AWS availability plus Amazon Bedrock Managed Agents powered by OpenAI, pointing toward enterprise-controlled agent deployments.

GitHub shows how production agent workflows reduced cost by logging token use, trimming unused MCP tools, and moving data access outside the LLM loop.

A practical look at where UI agents help, where they fail, and how accessibility surfaces can support robust automation beyond brittle pixel workflows.