Agentic AI
Amazon, Microsoft, Google converge on unified agent architecture—but portability remains locked to each cloud
Three hyperscalers have independently built nearly identical enterprise agent platforms with similar core layers, yet lack a neutral contract that would make agents truly portable across providers.
The New Stack
Within the past nine months, Amazon, Microsoft, and Google have each launched or rebranded an enterprise agent platform, and strikingly, all three share the same foundational architecture. Runtime, memory management, tool gateways, identity systems, observability, and governance now form the bedrock of Amazon Bedrock AgentCore, Microsoft Foundry, and the Gemini Enterprise Agent Platform, though each vendor uses different terminology. Where there was fragmentation across disparate libraries 18 months ago, a recognizable platform layer is now taking shape.
The parallel evolution of platform services#
This convergence mirrors what happened with Platform-as-a-Service between 2011 and 2016. Back then, developers stitched together virtual machines, load balancers, message queues, and monitoring tools—each with separate APIs and operational models. Tools like Cloud Foundry and Heroku unified these components around a standardized contract, allowing developers to think in terms of applications rather than infrastructure. Today's agent ecosystem sits at a similar inflection point, yet lacks an equivalent open contract to define interoperability.
The challenge is tangible: when a platform team deploys a customer support agent this quarter, they must select a model provider, choose a framework, provision memory storage, wire in a tool gateway for system integration, layer on identity controls, and eventually add observability. Each choice binds the agent to that vendor's ecosystem. Moving that agent later requires rebuilding the entire assembly—precisely the portability problem that PaaS solved for web applications a decade ago.
What the three platforms offer#
Amazon Bedrock AgentCore, launched in October 2025, provides seven composable services including session-isolated runtime with eight-hour execution windows, an API gateway that bridges existing MCP servers and Lambda functions, managed memory, code execution environments, identity isolation, and OpenTelemetry-based observability exported to CloudWatch.
Microsoft Foundry, renamed from Azure AI Foundry effective January 2026, delivers hosted agents in isolated managed runtimes, paired with Entra-based agent identity and multi-scoped memory (session, user, and procedural). Tracing also relies on OpenTelemetry.
