← Latest Update

Agentic infrastructure: models, orchestration, control, and credentials

Introducing Mistral Small 4. Mistral launches Mistral Small 4, a 119B Apache‑2 mixture‑of‑experts model that combines reasoning, multimodal inputs, and agentic coding abilities. This open‑weight MOE makes high‑end agent capabilities accessible for on‑prem and custom stacks—expect faster iteration on artifactized agents and new options for shipping production models (Principles 08, 09).

Use subagents and custom agents in Codex. Codex goes GA with subagents and TOML‑defined custom agents to run specialized, parallel, role‑based coding workflows. The pattern turns large agent jobs into orchestrated lanes you can test, version, and monitor—essential for building reliable delivery pipelines for outcome engineering (Principle 09).

Alibaba launches Wukong, an enterprise AI platform coordinating multiple agents (beta). Alibaba debuts Wukong, a beta platform for coordinating multiple agents to automate complex enterprise tasks like document editing. A vendor‑grade multi‑agent control surface changes where orchestration, governance, and enterprise integrations live—plan for service contracts, RBAC, and telemetry hooks when designing outcomes (Principles 09, 03).

Why agentic AI stalls in production — and how a control plane fixes it. The piece argues a control plane is necessary to coordinate agents, ground decisions in real‑time context, and restore observability and reliability in production. If you’re deploying agentic workflows, adding a control plane for routing, policy enforcement, and auditing is no longer optional but foundational to safe, debuggable outcomes (Principles 09, 06).

1Password introduces Unified Access platform and partner API for AI agent security. 1Password launches Unified Access to discover, secure, and audit credentials for both human users and AI agents, plus a partner API for integrations. Centralized credential discovery and enforcement closes a common threat vector as agents act across services—build credential lifecycle and proofing into your agent gate and audit processes (Principles 10, 15).