AWS unveils Loom: Amazon pushes its card on the control plane of enterprise agents

An open-source framework to govern AI agents at scale - a late but clear signal that the control plane is becoming a product.

In plain terms. AWS released Loom (InfoQ, 07/20/2026), an open-source reference platform for governing AI agents at the enterprise scale. Amazon's message to CIOs: the control plane for agents (access, audit, quotas, security) will be a market—and Amazon aims to be part of it. The move complements the open reading by Microsoft Build 2026 (#1063): the agent orchestration layer is becoming a standalone product.

Context

The enterprise agent market structured itself in 2025-26 around three layers: (1) the model, (2) orchestration (LangChain, LlamaIndex, Semantic Kernel), (3) deployment (Bedrock Agents, Azure AI Foundry, Vertex Agents). What was missing: the cross-cutting layer—who does what, with what rights, what audit, under what budget. This is where Loom positions itself, as an open-source counterpart to Copilot Studio.

The Data

• Publication: InfoQ, 07/20/2026 - "open-source reference platform."
• Announced scope: governance of agents at enterprise scale, integrations with the AWS stack (IAM, CloudTrail, Bedrock).
• Positioning: reference platform, not a managed product—code adaptable internally.
• Direct competitors: Salesforce Einstein Agent Studio, Google Agentspace, Microsoft Copilot Studio (closed/managed).

Analysis

Three things. One: Amazon acknowledges it is behind on the governance layer—Bedrock Agents remained rudimentary on RBAC and observability. Two: open source is a strategic admission—decision-makers do not want more lock-in on agents than on data. Three: this move makes Copilot Studio (closed, managed) more contestable for multi-cloud architectures, exactly when the "SaaS-is-over" thesis translates into agentic refactoring (see saas-is-over).

Scenarios

• Central (55%): Loom becomes a reference building block adopted by a fraction of Bedrock large accounts within 12 months, but remains marginal outside AWS.
• Cross-cloud adoption (25%): the cross-cutting model appeals to Azure/GCP customers; Loom becomes the de facto standard, following the Kubernetes 2015 model.
• Failure (20%): too early/too AWS-centric; supplanted by an emerging CNCF component in 2027.

Implications

For an IT decision-maker: do not sign an enterprise agent contract without a portability clause (RBAC, logs, policies)—Loom formalizes what needs to be negotiated. For SaaS agent providers (Sierra, Cognosys, Kognitos), the governance layer becomes mandatory, more differentiating. For investors: the enterprise agent stack shifts from "model-first" to "control-plane-first" in 2027.

To Watch

• GitHub adoption (stars/forks/enterprise deployment cases).
• Azure/GCP response within 90 days.
• Collision or complementarity with MCP (see mcp-ecosystem-plumbing).