🕷️
SetupAI
繁
💼Freshcollected in 4m

Enterprises Hedge Against AI Agent Control-Plane Risk

Enterprises Hedge Against AI Agent Control-Plane Risk
PostLinkedIn
💼Read original on VentureBeat
#agent-orchestration#cost-control#hybrid-cloud#observabilityvb-pulse-enterprise-ai-orchestration-researchmicrosoft ai foundrycopilot studioopenai agents sdkanthropicclaude agent sdk

💡See why enterprises are running three orchestration platforms—and still struggling to stop runaway agent costs.

⚡ 30-Second TL;DR

What Changed

85% of 107 surveyed enterprises use at least two orchestration tools, while 64% use three.

Why It Matters

AI teams should expect multi-platform agent architectures to become standard, increasing the need for portable workflows, unified observability, and vendor-neutral permissioning. Organizations that lack real-time cost controls may face unexpected spend as autonomous agents scale their tool and token usage.

What To Do Next

Create a platform inventory this quarter and enforce per-agent token budgets, real-time spend alerts, and an operator kill switch across Microsoft AI Foundry, OpenAI Agents SDK, and Anthropic deployments.

Who should care:Enterprise & Security Teams

Key Points

  • •85% of 107 surveyed enterprises use at least two orchestration tools, while 64% use three.
  • •Microsoft AI Foundry/Copilot Studio appears in 70% of enterprise stacks, followed by OpenAI Agents SDK at 68% and Anthropic Claude Platform at 47%.
  • •53% of respondents expect a hybrid primary control plane by the end of 2026.
  • •More than two-thirds plan to change orchestration platforms within 12 months, with Claude Agent SDK the leading option under consideration at 43%.

🧠 Deep Insight

Background and context from public sources — not the original article. 32 sources cited.

🔑 Enhanced Key Takeaways

  • •The concept of an "AI control plane" was formally recognized as an emerging market category by Forrester in December 2025, defined as enterprise infrastructure that inventories, governs, orchestrates, and assures heterogeneous AI agents across vendors and domains.
  • •Enterprise AI orchestration is crucial because, without it, AI assets (models, agents, data pipelines) operate in silos, leading to duplicated effort, inconsistent governance, and a gap between AI outputs and business processes.
  • •The cost of AI agent usage, particularly with platforms like Microsoft Copilot Studio, can vary significantly (e.g., 1 credit for a scripted FAQ vs. 100 credits for a reasoning-model response), making real-time cost visibility and management a major concern for enterprises.
  • •Shadow AI, where employees use unapproved AI tools, is identified as the largest unmanaged risk surface in most enterprises, with 78% of employees admitting to using such tools.
  • •The global hybrid AI deployment market, which balances cloud and on-premise AI, was valued at USD 43.00 billion in 2026 and is projected to reach USD 417.10 billion by 2035, growing at a CAGR of 28.72%.
📊 Competitor Analysis▸ Show
Feature/CategoryMicrosoft AI Foundry/Copilot StudioOpenAI Agents SDKAnthropic Claude Platform (Managed Agents)
Primary FunctionLow-code platform for building and deploying AI agents, integrated with Microsoft 365 and Azure services.Lightweight, Python-based framework for creating intelligent agents with instructions, tools, and delegation.Suite of composable APIs for building and deploying cloud-hosted agents at scale, handling infrastructure.
Pricing ModelComplex; often an add-on to M365 licenses. Internal agents for M365 Copilot-licensed users may not consume credits. External agents use prepaid capacity packs ($200/month for 25,000 credits) or pay-as-you-go ($0.01/credit). Credit consumption varies widely (1 credit for FAQ, 100 for reasoning).API pricing for underlying models (e.g., GPT-5.5: $5.00/million input tokens, $30.00/million output tokens). ChatGPT Enterprise is custom-priced.Standard Claude API token rates plus $0.08 per session-hour for Managed Agents. Claude Opus 4.8 API: $5.00/million input tokens, $25.00/million output tokens. Claude Enterprise starts from $20/seat/month plus usage.
Key FeaturesGoal-directed agents, attach tools (Azure AI Search, OpenAPI plugins, Logic Apps, Functions), multi-agent workflows, enterprise-grade security, observability, open standards (A2A, MCP).Agents (LLMs with instructions/tools), Handoffs (delegation), Guardrails (validation), Tracing. Supports sandbox, realtime, and voice agents. Python-first design.Secure sandboxing, long-running sessions, scoped permissions, tool execution, tracing, built-in orchestration harness. Multi-agent coordination and self-evaluation in research preview. Persistent memory.
IntegrationNative Microsoft Graph integration, Power Automate workflows, Azure services.Optimized for OpenAI models but designed to be provider-agnostic.Integrates with built-in tools like Bash, file operations, web search, and MCP servers.
LimitationsComplex cost estimation due to varied credit consumption and layered licensing.Intentionally avoids built-in graph-based workflow engine, vector memory, or opinionated agent planning system, shifting responsibility to developers for advanced behaviors.Multi-agent coordination and self-evaluation features are still in research preview.

🛠️ Technical Deep Dive

  • AI Control Plane Architecture: This architecture separates the control plane, which makes decisions and enforces policies (e.g., identity, routing, audit), from the data plane, which executes operations (e.g., agent runtime, model inference, tool execution). This separation is structural, akin to network engineering and Kubernetes.
  • Core Functions of an AI Control Plane: It is designed to inventory, monitor, enforce policy on, and audit every AI system across an enterprise, including LLM interactions, AI agents, Model Context Protocol (MCP) tool calls, and agent-to-agent communication.
  • Microsoft AI Foundry: This platform is designed for building, grounding, and governing AI applications and agents at scale. It includes an Agent Service, connected agents (A2A for agent-to-agent message exchange), and multi-agent workflows with a stateful layer for context, retries, and long-running steps. It emphasizes enterprise-grade security, observability, and open standards.
  • OpenAI Agents SDK: A Python-first framework built around a small set of primitives: Agents (LLMs equipped with instructions and tools), Handoffs (allowing agents to delegate tasks), Guardrails (for input/output validation and safety checks), and Tracing (for visualizing and debugging agent flows). It supports sandbox agents for isolated workspaces, realtime agents, and voice agents.
  • Anthropic Claude Managed Agents: This offering provides a managed infrastructure layer that handles the operational complexity of running AI agents at scale. It includes secure sandboxing, long-running sessions, scoped permissions, tool execution, and built-in tracing. The platform features an orchestration harness that manages tool calls, context, and error recovery, and supports multi-agent orchestration and persistent memory.

🔮 Future ImplicationsAI analysis grounded in cited sources

The role of CIOs will increasingly focus on establishing a unified Enterprise AI Control Plane (EACP) to manage AI sprawl and ensure governance.
As AI agent adoption outpaces governance, a centralized EACP becomes critical for unifying identity, policy enforcement, visibility, and cost control across diverse AI platforms and business units.
Hybrid AI deployment models will become the dominant strategy for enterprises, driven by the need for data residency, security, and cost optimization.
Enterprises are increasingly seeking a balance between deploying AI on cloud services and retaining sensitive data locally, with the hybrid AI deployment market projected for significant growth.
AI orchestration platforms will evolve towards providing more autonomous and self-improving workflows, reducing the need for constant human intervention.
The next frontier in AI workflow orchestration is workflow autonomy, with enterprises increasingly deploying autonomous AI teams, self-improving workflows, and dynamic model routing by 2026 and beyond.

⏳ Timeline

1950s
Turing Test Proposed, foundational for AI agents.
1980s
Intelligent Agents concept emerged, with AI systems operating with a degree of autonomy.
2023-Q4
Microsoft Power Virtual Agents rebranded to Microsoft Copilot Studio.
2025-03
OpenAI Agents SDK launched as a production-ready evolution of the experimental Swarm project.
2025-12
Forrester formally recognized the "agent control plane" as an emerging market category.
2026-04
Anthropic Claude Managed Agents launched in public beta.
💼Read original article on VentureBeat
📰

Weekly AI Recap

Read this week's curated digest of top AI events →

👉Related Updates

Same topic

Explore #agent-orchestration

Same product

More on vb-pulse-enterprise-ai-orchestration-research

Same source

Latest from VentureBeat

KCD Hangzhou Opens CFP for Agent Era Discussions

KCD Hangzhou Opens CFP for Agent Era Discussions

InfoQ中国•Aug 20
NanoClaw Brings Persistent Agent Teams to Slack

NanoClaw Brings Persistent Agent Teams to Slack

VentureBeat•Aug 20

AI-curated news aggregator. All content rights belong to original publishers.
Original source: VentureBeat ↗

This is a summary, not the original. Read the source, or get the weekly briefing.

Weekly AI briefing

One email a week. Unsubscribe anytime.