Agent Skills encodes senior-engineer workflows as injectable skills so agents generate specs, tests, and reviewable code instead of just “done”. That gives you a practical pattern to push intent, quality, and test scaffolding into agent outputs so reviewers and downstream systems can rely on deterministic artifacts.

The RAG era is ending for agentic AI — a new compilation-stage knowledge layer is what comes next launches Nexus and KnowQL, a compilation-stage knowledge engine that precomputes task-ready artifacts and provides deterministic, auditable retrieval. That reframes context engineering: invest in precompiled, testable knowledge artifacts to make agent behavior reproducible and easier to validate.

Microsoft takes Agent 365 out of preview as shadow AI becomes an enterprise threat announces GA for Agent 365 to discover, govern, and secure employee AI agents. If you run agents in enterprise settings this elevates inventory, policy enforcement, and discovery to core infrastructure required to manage risk and shadow AI.

Inside Amex’s agentic commerce stack: How intent contracts and single-use tokens enforce AI transactions describes Amex’s ACE developer kit centralizing transaction validation with intent contracts and single-use payment tokens. That provides a concrete pattern for locking down side-effects from agents — transaction-level validation and single-use tokens give you auditable gates for real money and compliance-sensitive actions.

IBM charts AI operating model to move enterprises beyond experimentation unveils an AI operating model aimed at scaling agent orchestration, hybrid-cloud data integration, and digital sovereignty for measurable ROI. Treat this as a reference architecture: it codifies roles, runtime orchestration, and integration patterns you need to move agents from pilots into reliable production.