Summary Bullets:

• Fast-growing use of agentic AIs within organizations has triggered agentic orchestration/governance prioritization among platform providers

• Controversy remains over two distinct approaches to orchestration: control plane construct or orchestration frameworks

Enterprises deploying AI in 2026 are turning their attention from deployment of agentic AIs to the management of growing numbers of agents being released across organizations. Companies are struggling with how to manage the hundreds or thousands of individual agents built within their organizations–agents built by different teams, running on different platforms, with inconsistent security and governance. This is problematic, considering most organizations lack visibility into agent inventory, purpose, and authorization.

The practice of addressing agentic orchestration at the control plane platform layer is moving to the forefront of the conversation, spurred by the lack of visibility, security, and management associated with agentic sprawl. Control planes sit above the agent layer, governing, observing, and enforcing policy across agents regardless of where they originated. The advantage is in their ability to ensure identity and security enforcement and enable cross-vendor interoperability regardless of what framework they use for coordinating agents. This approach contrasts with orchestration frameworks which are features of agentic solutions that simply coordinate agents.

GlobalData’s research captures the scale of the agentic market, and therefore the urgency of the situation. The report “Market Opportunity Forecasts to 2029: Agentic AI” puts the global agentic AI market at a 50.6% CAGR for 2024–2029, reaching $45.4 billion by 2029, driven by enterprise demand for autonomous decision-making, multi-agent orchestration, and scalable cloud-native AI infrastructure. GlobalData reports that early adopters are even replacing traditional robotic process automation (RPA) with goal-driven, self-adapting agent systems — and that the shift from pilots to production-grade systems is accelerating.

Vendor Strategies

A control plane market is emerging, positioned as framework-agnostic. Leading AI and platform providers are announcing strategies and solutions to address this evolving branch of agent orchestration:

• IBM is positioning the next generation of watsonx Orchestrate as an agentic control plane for the multi-agent era. It supports IBM-native agents alongside LangGraph, Langflow, and agents built on the open A2A protocol, with consistent policy enforcement.

• Salesforce has built its orchestration strategy on MuleSoft’s Agent Fabric. This has been helped by its ability to consolidate multiple data sources into a single source following Salesforce’s Informatica acquisition last November. A trust and data security layer serves as the key component of its new Agent Fabric control plane.

• ServiceNow is featuring its AI Control Tower as the governance layer spanning every AI agent, model, and action running across the enterprise, regardless of which vendor built them. The company is repositioning from being a workflow automation vendor to an enterprise AI operating system, shored up by its recent acquisition of IT/OT security provider Armis, which leans heavily into its new AI Control Tower solution.

• Boomi’s control plane approach is addressed via the Boomi Enterprise Platform, which sits between disparate systems, agents, frontier models, and data sources. Boomi’s acquisition of Lunar.dev, AI/MCP gateway, plays heavily into its strategy as the prompt routing layer for governing MCP servers and access.

Yet controversy over how to govern the fast-growing agentic AI market segment remains. Some rival platform providers are taking a different tact and keeping agentic orchestration within the confines of their own platforms and product ecosystems. They are not positioned as supporting cross-vendor governance layers in the same way as competitors:

• Microsoft has been reshaping Copilot Studio from an agent-building tool into an agent governance layer. It describes the new governance features as having centralized policy enforcement, agent lifecycle oversight, and cross-ecosystem governance spanning Microsoft 365 and partner-built agents. However, Microsoft’s architecture is embedded in and distributed across its popular platforms, including Power Platform and Azure, versus a discrete, specific control plane layer that sits above disparate agents.

• Oracle OCI’s strategy for management and governance also currently bypasses a control plane architectural model and remains within the confines of its own ecosystem. OCI Enterprise AI embeds agentic orchestration natively as a feature across the Oracle technology layers rather than positioning a discrete governance layer above them. Enterprise AI’s three integration layers are: Enterprise AI Models, Enterprise AI Agents, and Enterprise AI Governance.

Summary
The concept of a control plane architecture construct is still being defined by the market. Vendors operating in the agentic orchestration space have varying opinions and product strategies. Pioneering activities and offerings suggest this type of AI operating system will quickly become the fundamental layer for agentic AI. Operational guardrails are critical for bringing to production environments that are built around ambitious agentic AI projects.

It is worth noting that players in this market segment have generally adopted or endorsed MCP and A2A as the underlying interoperability layer, serving as the common protocol layer, while the control planes above it remain proprietary and competitive. Therefore, much of the agentic AI battle will be won or lost according to who controls the management, orchestration, and governance of disparate agents across enterprise environments.

For more on this topic and other cloud trends including escalating cloud costs, please see Cloud Watch Q2 2026: Reassessing On-Demand Economics in the Era of Escalating Cloud Costs

The post Lack of AI Agent Oversight Brings Dueling Approaches appeared first on IT Connection.

Summary Bullets:

• Cycles between advanced AI model rollouts are significantly shortened among leaders in this space

• Developers are gaining access to agentic-injected integrated development environments (IDEs); while knowledge workers gain access to agentic AI assistants.

The second quarter marks a momentous period in the industry’s ongoing AI efforts. Platform leaders shipped next-generation agentic runtimes including autonomous and other advanced capabilities, all while managing a more compressed cycle of new AI models, which are rolling out in a matter of weeks versus months.

A few notable announcements highlight this structural shift in how enterprise AI is built, deployed, and presented to enterprises.

Microsoft’s long-awaited private review of its first in-house reasoning model, MAI-Thinking-1, an enterprise-grade medium-weight model that promises to shake up the industry in a number of ways. Microsoft is going up against the industry’s strongest models based on the strength of its mathematical and scientific reasoning abilities, for improved training loops, citing numerous Microsoft-backed engineering benchmark tests. It is taking on Claude Sonnet 4.6 and Opus 4.6 by claiming lower token costs and smaller inference footprint. For the first time since the beginning of its relationship with OpenAI, Microsoft is able to break into the enterprise space with its own AI model, on par with leading rivals. Microsoft’s win will inevitably be at the expense of OpenAI.

Expanding its AI portfolio further was the June release of Microsoft Copilot Studio – Computer Use, revamping AI assistants to perform further up the agentic AI stack. The release supports the use of computer-use agents directly in Copilot Studio, helping bypass integrations with APIs in order to develop workflow automations.

To keep pace with top rivals Google and Anthropic, OpenAI announced its biggest model release yet, GPT-5.5, emphasizing its strengths in agentic coding, scientific research, and the ability to automate tasks associated with knowledge work. As the industry’s early GenAI leader, OpenAI has been challenged to maintain its innovative prowess.

OpenAI’s newest advancements are mere weeks following its last GPT release, demonstrating the staggering breakneck pace AI model providers are compelled to maintain to keep up in this highly competitive segment. OpenAI is hoping to win back the loyalty of professional coders who have moved to Anthropic Claude in droves for its accuracy in coding. OpenAI has been most popular among consumers, while competitors, including Google and Anthropic, have gained more traction in the enterprise space.

AWS’s latest AI announcements demonstrate a deliberate pivot towards agentic AI amidst an increasingly competitive landscape. Under mounting competitive pressure, Amazon is investing heavily in tools that span developer and non-developer audiences.

The newly announced Amazon Quick agentic AI assistant is a revamp of the GenAI assistant Q Business platform, providing knowledge-based workers with insights while also being able to act and automate repetitive workflows. Quick connects internal data across AWS services, third-party platforms, and on-premises systems. Other key announcements were Kiro agentic IDE built on Code OSS and powered by Claude models, via Amazon Bedrock; and Bedrock AgentCore, serverless runtime, and AgentCore Harness, which let developers build and run production-grade AI agents quickly without needing to code custom orchestration loops.

The post AI Wars Intensify via Major LLM/Agentic Releases appeared first on IT Connection.