A July 20 analysis from The New Stack suggests that Amazon, Microsoft, and Google appear to be converging toward similar architecture patterns. For enterprise buyers, the significance is not that a formal cross-cloud standard exists, but that the three largest platform vendors now appear to be assembling the same core stack components while keeping the ability to move workloads between them unresolved.
What changed
The immediate news hook is analytical, not a vendor-announced standard. In a July 20, 2026 article, The New Stack analysis indicates that Amazon, Microsoft, and Google have converged on a shared enterprise agent architecture and that the agent ecosystem is becoming a distinct platform layer.
The milestones cited as evidence are recent and concrete. Amazon Bedrock AgentCore reached general availability in October 2025 with seven composable services. Microsoft renamed Azure AI Foundry to Microsoft Foundry effective January 1, 2026. Google retired the standalone Vertex AI product line at Cloud Next 2026.
The New Stack’s architecture thesis identifies six recurring components across the three platforms: runtime, memory, tool gateway, identity, observability, and governance.
Why B2B teams should care
The main operational issue is portability. The article concludes that identity, telemetry, and deployment terminate within one provider, which makes agents hard to move even if the high-level architecture looks similar across clouds.
BCG’s framework complements this discussion by outlining the architectural capabilities enterprises should standardize internally. BCG says companies need interoperability, security, monitoring, and governance across teams and use cases, and it argues that a common platform should standardize orchestration, model access, evaluations, guardrails, memory, security, and monitoring.
BCG’s three-layer framing is also useful as an evaluation lens: semantic intelligence, data abstraction, and governance and operations.
Who is affected
Large enterprises are the clearest audience, especially in regulated industries such as health care and banking. BCG says regulated sectors need clearer architecture and explicit controls around identity, access, and entitlements. Separately, BCG says agentic systems introduce multiagent orchestration, agent hierarchies, shared intent and memory, and mixed deterministic and nondeterministic workflows.
The governance gap appears wider than deployment numbers suggest. According to a 2026 OutSystems survey cited by TechTimes, 96% of enterprises already run AI agents in production, but only 12% say they can govern them. The same report claims 97% of organizations are exploring agentic AI strategies, while only 36% have a centralized governance approach.
What teams should check now
Platform evaluations should move past vendor demos and into architecture reviews. Based on the shared stack components identified in the source pack, teams should inspect:
- model access and routing
- tool calling and gateway design
- identity and permissions
- memory handling
- orchestration
- observability and telemetry
- governance controls
- auditability
AWS: According to InfoQ, AWS positions Loom as a concrete inspection checklist and an example of how organizations might build their own agent platforms. Teams assessing AWS should verify identity provider integration, scope-based authorization, lifecycle management, RFC 8693 token exchange through AgentCore Identity, mandatory resource tagging, and integration with AWS Agent Registry, which is currently in public preview.
Microsoft: Authentication boundaries need similar scrutiny on Microsoft’s side. While unrelated to agent platforms directly, a July 20 Microsoft post about the new Fabric link experience illustrates Microsoft’s current identity direction: workspace identity is the recommended authentication option. That post is not an agent-platform launch, but it is a useful control-plane signal: teams should verify where authentication terminates and whether workspace identity or provider-native identity becomes the practical point of control.
Network: Network and telemetry dependencies need similar mapping. A CIO analysis says real-time observability and telemetry are critical requirements. For enterprise agent deployments this matters because future-ready networks need end-to-end visibility across campus, branch, cloud, and data center environments.
What remains unclear
- Not yet confirmed: whether any primary Microsoft architecture document defines the same runtime, memory, tool gateway, identity, observability, and governance contract that The New Stack describes across all three vendors.
- Not yet confirmed: whether any primary Google architecture document establishes a single common cross-vendor contract rather than a platform-specific design.
- Not yet confirmed: the full contract mapping in The New Stack article, because the available excerpt is truncated before that section is fully shown.
- Not yet confirmed: any direct vendor quotes in the scraped New Stack excerpt supporting or disputing the convergence thesis.
- Not yet confirmed: a formal standard across AWS, Microsoft, and Google, since the New Stack piece is an analysis/opinion article rather than a joint vendor announcement.
What to watch next
The next signal is whether any vendor or open-source project defines a portability contract for enterprise agent architecture. The New Stack reports that no open-source project has defined an equivalent agent contract.
A second signal is whether agent gateways emerge as the control plane, an argument Forbes contributor Janakiram MSV has made and that MarketScale cites in its own analysis, which also reports that AWS is among the vendors building agent gateway capabilities.
Sources
- The New Stack, Amazon, Microsoft, and Google are converging on the same enterprise agent architecture
- BCG, Building Enterprise AI Agents in Regulated Industries
- TechTimes, Gemini Enterprise Agent Platform leads enterprise AI governance; OpenAI starts billing agents
- Microsoft, Link to Fabric UX refresh
- CIO, Why the future of AI depends on the network.
- InfoQ, AWS recently released Loom, an open-source, opinionated agent platform…
- MarketScale, Enterprise AI’s infrastructure moment: agents, gateways, and the battle for the control plane
