What's shipping in enterprise AI.

NTT DATA Bets on Gemini Enterprise: 5,000 Experts and 500 Agents Signal the SI Alliance Race (opens in a new tab)

NTT DATA's June 8 expansion with Google Cloud — 5,000 certified Gemini Enterprise experts and up to 500 co-deployed agents — marks the opening move in what will become a systems integrator alliance race across the three major model vendors.

ServiceNow AI Control Tower Expands Across Microsoft and Third-Party Agent Estates (opens in a new tab)

ServiceNow used Knowledge 2026 to extend AI Control Tower across Microsoft Agent 365 and launch Autonomous Security & Risk with Armis and Veza. A week later, Cognizant layered real-time Guardian agents on top — the first commercial implementation of continuous AI assurance as platform infrastructure rather than periodic audit.

Mem0 new memory algorithm: +29.6 points on temporal queries, +23.1 on multi-hop at 73% lower token cost (opens in a new tab)

Mem0 released a new token-efficient memory algorithm (April 2026) showing gains of +29.6 points on temporal queries and +23.1 points on multi-hop reasoning over their prior ECAI 2025 baseline on the LoCoMo benchmark. Token cost per retrieval dropped from ~26,000 to ~6,956 tokens — a 73% reduction. The algorithm treats agent-generated facts as first-class alongside user-stated facts. Separate MemoryArena benchmarks quantified the cost of not managing memory: task completion dropped from 80%+ to ~45% when replacing an active memory agent with long-context-only.

• Agent Memory
• ·Memory Consolidation
• ·Retrieval / RAG

EU AI Act high-risk enforcement activates August 2, 2026 — 60 days for agentic AI compliance (opens in a new tab)

EU AI Act Articles 8–17, 26, and 73 — the high-risk obligations covering risk management, human oversight, audit logging, technical documentation, and serious incident reporting — activate August 2, 2026. Enterprise agentic AI deployments in European regulated industries (financial services, insurance, healthcare) are directly in scope. The compliance requirements most difficult for agentic systems to satisfy are: per-agent identity, comprehensive audit trails, mandatory human-sign-off gates for consequential actions, least-privilege tool scoping, and vendor-supplied technical documentation. California SB 1047 was stayed and replaced with narrower SB 189.

• Governance
• ·Autonomous Agents
• ·Agent Orchestration

Microsoft Build 2026: Agent Framework GA, Foundry Hosted Agents, and the Agent Control Specification (opens in a new tab)

Microsoft Build 2026 (June 2–3, San Francisco) delivered four major agentic AI moves: Microsoft Agent Framework 1.0 reached GA as the formal merger of AutoGen and Semantic Kernel into a single .NET/Python SDK; Azure AI Foundry Hosted Agents reached GA with hypervisor-isolated per-agent sandboxes, per-agent Entra ID identities, sub-100ms cold starts, and zero idle cost; the Agent Control Specification (ACS) was published as an open governance spec for agent runtime behaviour; and Web IQ / Work IQ / Foundry IQ were shipped as a unified governed context layer for agents. A2A (agent-to-agent) protocol and full MCP support are built in, enabling cross-provider interoperability with Claude and Gemini agent stacks.

• Agent Orchestration
• ·MCP
• ·Autonomous Agents
• ·Governance

Snowflake Summit 2026: Agent Identity, Horizon Context, and the Natoma MCP Acquisition (opens in a new tab)

Snowflake's Summit 2026 launched 26+ capabilities anchored by three governance moves that matter for enterprise AI: cryptographic agent identity at GA, the Horizon Context semantic metadata layer, and the acquisition of Natoma — an enterprise MCP gateway that enforces policy at the tool-call boundary.

Skan AI introduces Agentic Business Context Foundation (opens in a new tab)

Skan AI, an enterprise context graph company, introduced the Agentic Business Context Foundation (ABCF), a framework that captures context traditional enterprise systems miss — human reasoning, exceptions, and informal workarounds — and converts it into actionable context for AI agents. The framing is that agents trained only on documentation and event logs falter at the edges where high-value work lives.

• Context Graph
• ·Enterprise Semantics
• ·Agent Memory

AWS MCP Server reaches GA with full API coverage and IAM-based governance (opens in a new tab)

AWS has made its managed Model Context Protocol server generally available, giving AI coding agents controlled access to AWS APIs, documentation, and operational workflows through a standard interface. The MCP server provides IAM-based access controls, CloudWatch metrics, and CloudTrail logging, enabling organizations to govern and audit agent activity independently.

• MCP
• ·Governance
• ·Agent Orchestration

MCP 2026-07-28 Spec RC: Stateless Core and Auth Hardening Remove the Last Enterprise Barriers (opens in a new tab)

The MCP 2026-07-28 release candidate, published May 21, is the protocol's largest revision since launch. A stateless HTTP core eliminates the scaling friction that kept enterprise MCP deployments small; OAuth/OIDC-aligned auth removes the custom wrapper overhead that compliance teams have been absorbing. Final spec publishes July 28.

• MCP
• ·Agent Orchestration
• ·Governance
• ·Autonomous Agents

AllegroGraph v9 ships GraphTalker, an agentic NL interface for enterprise knowledge graphs (opens in a new tab)

Franz Inc. released AllegroGraph v9 with GraphTalker, a natural-language AI agent that lets business users, data scientists, and graph engineers query enterprise knowledge graphs without writing SPARQL. GraphTalker uses an iterative agentic loop — inspect schema, review examples, generate queries, test, learn from errors, refine — rather than one-shot prompt-to-query translation.

• Knowledge Graph
• ·Ontology
• ·Autonomous Agents

NSA publishes security design considerations for MCP-based AI automation (opens in a new tab)

The NSA Artificial Intelligence Security Center released a Cybersecurity Information Sheet, 'Model Context Protocol (MCP): Security Design Considerations,' setting out guidance for AI-driven automation built on MCP. The guidance covers parameter validation, tool segregation by data classification zone, and a preference for local MCP server instances when processing private data.

• MCP
• ·Governance

Redis launches Context Engine for AI agent memory (opens in a new tab)

Redis has unveiled the Context Engine, a dedicated memory layer for enterprise AI agents composed of Redis Context Retriever, Redis Agent Memory, and Redis Data Integration. Redis Data Integration is generally available from May 18, 2026; the Context Retriever and Agent Memory components are in preview. Redis frames it as solving the context problem that causes autonomous agents to hallucinate or stall.

• Agent Memory
• ·Retrieval / RAG
• ·Vector Search

SAP Business AI Platform and AI Agent Hub: The Autonomous Enterprise Governance Stack Arrives (opens in a new tab)

SAP Sapphire 2026 delivered the most architecturally significant enterprise AI announcement of the quarter: a three-layer Business AI Platform consolidating BTP, Business Data Cloud, and Business AI — with a governance command center (AI Agent Hub, Q3 GA) that manages SAP and non-SAP agents from a single plane.

• Agent Orchestration

SAP unveils the autonomous enterprise with SAP Knowledge Graph (opens in a new tab)

At SAP Sapphire 2026, SAP introduced SAP Knowledge Graph as the foundational context layer for the autonomous enterprise, alongside SAP-RPT-1.5 for tabular reasoning and SAP Autonomous Suite with over 50 domain-specific Joule Assistants orchestrating 200+ specialized agents. SAP also announced the intent to acquire Reltio to unify master data.

• Knowledge Graph
• ·Agent Orchestration
• ·Autonomous Agents
• ·Enterprise Semantics

Anthropic ships Dreaming primitive for asynchronous agent memory consolidation (opens in a new tab)

Anthropic shipped its Dreaming primitive on May 6, 2026 — an asynchronous process that runs between agent sessions, reviewing session transcripts and memory stores, extracting patterns, and merging duplicates. It is part of a four-development cluster between April 23 and May 24, 2026 that has moved agent memory from a framework-level concern to a platform-level primitive on the major model APIs.

• Agent Memory
• ·Memory Consolidation
• ·Autonomous Agents

Oracle introduces governed memory core for enterprise AI agents (opens in a new tab)

Oracle has released Oracle AI Agent Memory, a model- and framework-agnostic Python package built on Oracle AI Database, providing short-term threads with summaries and context cards, long-term durable memories with vector search, automatic LLM-based memory extraction, and the governance and isolation production agents require.

• Agent Memory
• ·Vector Search
• ·Governance

AWS Bedrock AgentCore GA: Managed Agent Infrastructure Reaches Enterprise Production (opens in a new tab)

Amazon Bedrock AgentCore reached general availability in late April 2026, delivering the first fully managed, framework-agnostic agent infrastructure on AWS — covering runtime, memory, gateway, identity, and observability without requiring teams to build or maintain the underlying platform.

• Autonomous Agents
• ·Agent Orchestration
• ·Governance
• ·MCP