Notion opens its workspace to external coding and service agents via External Agents API and Workers

What changed: Notion’s Developer Platform now surfaces third‑party agents (Claude Code, Cursor, OpenAI Codex, and a customer‑service agent) as tracked collaborators inside a workspace and launched Workers, a hosted runtime for custom code; Workers has a free preview through August 11, 2026, and the External Agents API is in private beta.

Why it matters: Teams can assign tasks to best‑of‑breed coding or service agents without leaving Notion, keep results and context in the same workspace, and run small integration logic without separate infrastructure — lowering friction for building agent‑driven automations and reducing context‑switching for knowledge work.

Try/watch: Join the External Agents API waitlist and use the Workers free preview to prototype a small workflow (for example: ticket → agent debug → human review → database update). Monitor access controls and the documented prompt injection mitigations as agents gain access to broader workspace data.

GitHub Agentic Workflows ships v0.75.4 — Codex hardening, explicit permission mode, and better observability

What changed: GitHub’s Agentic Workflows project published a May 25 weekly update and v0.75.4 pre‑release that hardens the Codex engine, adds explicit engine.permission‑mode to make tool permissioning auditable, improves OpenTelemetry trace inheritance for child processes, fixes Gemini stream parsing, and sets the Codex default model to gpt-5.3-codex.

Why it matters: Builders deploying coding agents or automated developer workflows get more stable execution, clearer security boundaries (so agents can’t silently bypass allowed‑tool checks), and better end‑to‑end observability — all practical improvements that reduce risk and troubleshooting time when scaling agentic automation.

Try/watch: Upgrade your staging environments to v0.75.4, test engine.permission‑mode settings to enforce least‑privilege for agent tools, and validate distributed traces with OpenTelemetry to ensure you can reconstruct agent decisions and failures during post‑deployment audits.