The three platforms, side by side

AWS — Bedrock AgentCore

The shape. AgentCore is not one service; it is a set of modular services that work together or independently — Runtime, Identity, Memory, Gateway, Policy, built-in sandboxed tools (Code Interpreter, Browser), Observability, and Evaluations, with a separate Agent Registry for cataloguing agents and tools. It is deliberately framework- and model-agnostic: it hosts CrewAI, LangGraph, LlamaIndex, or AWS’s own Strands SDK, with models on Bedrock or elsewhere. AgentCore reached general availability on 13 October 2025. If you want no orchestration code at all, the managed harness (GA July 2026) lets you declare model, tools, and instructions instead.

Runtime story. Serverless, session-first. Each session is keyed by a runtimeSessionId and runs in a dedicated micro-VM with isolated CPU, memory, and filesystem; when the session ends the whole micro-VM is destroyed and its memory sanitised, so cross-session contamination is prevented by the substrate rather than by your code. Sessions are ephemeral by design — anything durable belongs in AgentCore Memory. A second substrate, Runtime Instances (GA August 2026), runs sessions on EC2 instance families with persistent volumes for much longer-lived work.

Tool story. MCP everywhere, via Gateway. Gateway turns Lambda functions, API Gateway stages, OpenAPI schemas, Smithy models, and remote MCP servers into MCP tools; for MCP targets it runs in aggregation mode, presenting itself as one MCP server whose tools/list consolidates every attached target. Because a hundred tools would drown the context window, Gateway embeds each tool’s name, description, and parameter descriptions at sync time and exposes semantic tool search as a built-in MCP tool. Gateway is also the enforcement point: AWS positions it as providing both ingress auth (who may call the gateway) and egress auth (credentials for each backend).

Identity story. AgentCore Identity treats agents as first-class workload identities, each with its own ARN in a central agent identity directory. Inbound, you choose IAM SigV4 for callers inside your AWS boundary or OAuth JWT bearer tokens validated against an IdP you configure (Cognito, Okta, Entra ID) at runtime creation. Outbound, a token vault stores OAuth access and refresh tokens, API keys, and client secrets encrypted with KMS keys, so the agent acts on behalf of a specific user without ever holding a shared secret.

Microsoft — Foundry Agent Service

The shape. Microsoft Foundry Agent Service offers exactly two agent types. Prompt agents are declarative: instructions, model, and tools as configuration, with no containers or packages to maintain, authored either in the portal or through SDKs for CI/CD. Hosted agents are bring-your-own-code — a container image or a .zip of source Foundry builds for you — written with Microsoft Agent Framework, LangGraph, the OpenAI Agents SDK, the Anthropic Agent SDK, the GitHub Copilot SDK, or plain custom code. A third pattern exists but is not a persisted agent: calling the Responses API directly from your own app creates an ephemeral agent whose definition lives in your code.

Runtime story. The runtime rests on three primitives: agents (persisted, automatically versioned definitions), conversations (durable server-side history of items — messages, tool calls, tool outputs), and responses (the unit of execution). Hosted agents run in per-session VM-isolated sandboxes with a persistent filesystem ($HOME and /files), which is what makes scale-to-zero with stateful resume possible: the platform deprovisions compute at an idle timeout you configure between 2 and 60 minutes (default 15), then restores state when the session wakes. Sessions are deleted after 30 days of inactivity. A separate durable key-value state store holds keyed JSON — including framework checkpoints — independently of compute.

Tool story. A Toolbox groups curated tools — web search, file search, code interpreter, MCP servers, OpenAPI tools, custom functions — behind one managed MCP-compatible endpoint, centralising authentication, governance, and versioning; you can promote a new toolbox version to default without redeploying agents. Because the endpoint is MCP, it is Foundry-homed, not Foundry-bound: Agent Framework, LangGraph, or any MCP client can consume it. Tool search (preview as of September 2026) hides tools by default and exposes only two meta-tools, tool_search and call_tool, so a toolbox can hold hundreds of tools without flooding the model’s context.

Identity story. Microsoft Entra Agent ID extends Entra to agents: an agent identity is a special kind of service principal with an object ID, a display name, and an optional human sponsor who is accountable for it, created from a reusable blueprint that acquires tokens on its behalf. In Agent Service, each hosted agent automatically receives a dedicated Entra identity; when a user token is present the agent uses OAuth 2.0 On-Behalf-Of to call downstream services with delegated permissions, and without one it authenticates as itself. Launch stages here have moved recently — at least one Entra Agent ID page still carried a preview banner in September 2026, so verify before you write it into a control document.

Google — Agent Platform (formerly Vertex AI)

The shape — and the caveat. Google’s agent stack is mid-rebrand, so hold every name loosely. As of September 2026 the product page is titled “Gemini Enterprise Agent Platform (formerly Vertex AI)”, the Vertex AI documentation carries a banner saying it is no longer being updated, and the managed runtime documented for a year as Vertex AI Agent Engine is now presented as Agent Runtime. Separately, ADK — the open-source Agent Development Kit, available in Python, TypeScript, Go, Java, and Kotlin — is the build side, and the Gemini Enterprise app is the employee-facing agent surface. The rebrand’s announcement date was not established when this pack was checked, so treat 2026-09 as “observed by”, not “changed on”.

Runtime story. Agent Runtime is described as a fully managed environment handling testing, release management, and reliability so you deploy and scale agents without managing infrastructure, and it hosts agents built with ADK, A2A, LangChain, LangGraph, AG2, LlamaIndex, or fully custom code. Around it sit managed Sessions (individual interactions between users and agents), Memory Bank, Example Store, evaluation, a secure Sandbox for code execution — including Computer Use and shell sandboxes — and a Feedback service. Operational features documented include Terraform provisioning, revision and traffic management, Cloud Trace, Cloud Logging, bidirectional streaming, and Private Service Connect interfaces. The overview page does not print launch stages, so check each feature’s stage yourself.

Tool story. Tools are mostly an ADK concern — function tools, MCP tools, OpenAPI tools, and built-ins such as Google Search, with tool authentication support. At the platform layer, Agent Gateway (marketed as “Agent-to-Anywhere”) handles routing and authorization for calls made during agent execution, and a Skill Registry stores, searches, and dynamically loads agent skills. ADK also has first-class A2A support for exposing and consuming remote agents — unsurprising, since Google authored A2A before donating it to the Linux Foundation.

Identity and memory story. Identity is IAM-based: agents run under service accounts you control, with VPC Service Controls, CMEK, data residency at rest, and Access Transparency documented per service in an enterprise-security matrix — and not uniformly across services, so re-read that matrix rather than assuming a feature covers everything. Memory splits the same way as its rivals: Sessions hold short-term conversational state, while Memory Bank uses generative models to generate long-term memories from session data rather than storing turns verbatim.

What happened to Amazon Bedrock Agents — the 2023 one?

It became Amazon Bedrock Agents Classic and closed to new customers on 30 July 2026. Existing customers keep it in maintenance mode: no new features, but AWS explicitly states there is no planned end-of-life date and no migration deadline. Do not tell your stakeholders it is being shut down — that claim is not in the docs.

Bedrock itself, Knowledge Bases, and Guardrails are unaffected. The documented migration paths are the AgentCore harness (declare model, tools, instructions; AgentCore supplies compute, memory, identity, observability) or code-defined agents on AgentCore Runtime for custom orchestration. Old action groups become MCP tools behind Gateway or code-level tools; AMAZON.CodeInterpreter becomes AgentCore Code Interpreter; session config becomes Memory strategies.

What happened to the Azure agents with threads, runs, and messages?

That Assistants-style model is Agents (classic), now deprecated with retirement on 31 March 2027; the Azure OpenAI Assistants API sunset on 26 August 2026, along with the azure-ai-inference package. The new model maps Assistants → Responses API, and Threads/Messages/Runs/Assistants → Conversations/Items/Responses/Agent Versions. Agents are now identified by name plus version rather than a GUID.

Practical consequence for anything you read online: tutorials that create a thread and poll a run are describing the retiring service.

Are Foundry’s visual workflows a safe place to build multi-agent orchestration?

No — and this is the clearest “don’t learn the doomed thing” warning in the pack. Foundry’s visual workflows are in preview and retire on 1 December 2026; after that the designer and in-portal execution are unsupported. Microsoft’s recommended replacements are Microsoft Agent Framework (code-first or declarative YAML), Azure Logic Apps, or direct A2A connections between agents. Exported workflow YAML remains runnable when deployed as a hosted agent.

If you need multi-agent patterns on Azure, learn Agent Framework’s orchestrations — Sequential, Concurrent, Handoff, Group Chat, and Magentic — not the designer.

Why does Google’s documentation contradict itself about names?

Because three renames overlapped. “Agent Builder” has meant at least three things: the 2024 search-and-conversation product (renamed AI Applications in April 2025), the 2025 suite of agent-building features, and now a brand folded into Gemini Enterprise Agent Platform. Meanwhile Vertex AI Agent Engine (GA March 2025, itself renamed from “LangChain on Vertex AI”) is presented as Agent Runtime, and Vertex AI Extensions was deprecated in May 2026 with shutdown after 26 November 2026.

Survival tactic: navigate by capability — “the managed runtime”, “the memory service”, “the tool gateway” — and confirm the current product name and launch stage on the live page before you cite it.

And the frameworks — Semantic Kernel, AutoGen, Strands, ADK?

Each cloud now has a first-party open-source SDK, and none of them is required by its cloud’s platform. Microsoft Agent Framework (announced 1 October 2025) is the documented next generation of both Semantic Kernel and AutoGen, which remain supported while investment moves. Strands Agents is AWS’s open-source, model-driven SDK for Python and TypeScript — one of several frameworks AgentCore hosts, not a prerequisite. ADK is Google’s, in five languages, with ADK 2.0 adding graph workflows.

The important structural fact: framework choice and hosting choice are separate decisions. That is the whole point of the crosswalk to the framework landscape module.

Managed agent platforms on the cloud timeline

• 2023-11-01 — Amazon Bedrock Agents launches:

  The first major cloud-managed agent service arrived in November 2023: action groups backed by Lambda, a knowledge-base integration, and a hosted orchestration loop. It marks the point where agents became infrastructure to procure, not just code to write — and it is now “Agents Classic” in maintenance mode, which is its own lesson about building on managed abstractions.

• 2025-03-04 — Vertex AI Agent Engine reaches GA:

  Google’s managed agent runtime — previously “LangChain on Vertex AI” — went generally available with billing starting the same day. That is sixteen months after Amazon Bedrock Agents made a managed agent loop something you procure rather than write.

• 2025-04-09 — Google ADK launches in preview; Agent Builder becomes a suite:

  The open-source Agent Development Kit arrived in preview at the same event, with Agent Engine gaining ADK integration and an Example Store, Agent Garden debuting, and the original Agent Builder renamed AI Applications. ADK Python reached a production-ready 1.0.0 at I/O on 20 May 2025, alongside a Java SDK.

• 2025-05-01 — Azure AI Foundry Agent Service reaches GA:

  Microsoft’s managed agent service went generally available around Build 2025 with connected agents, agent tracing, and Logic Apps triggers. Everything about it — the name, the API, the threads-and-runs model — changes at least twice in the eighteen months that follow.

• 2025-07-16 — AWS previews Amazon Bedrock AgentCore:

  Seven services in four regions — Runtime, Memory, Gateway, Browser Tool, Code Interpreter, Observability, Identity — sold as primitives rather than one opinionated agent object. This is AWS conceding that its 2023 Bedrock Agents abstraction was the wrong shape.

• 2025-10-10 — Gemini Enterprise becomes Google’s agent front door:

  Announced as “the new front door for AI in the workplace”: Gemini models, a no-code workbench, prebuilt agents, connectors, and governance in one product, succeeding Agentspace. The pattern across all three clouds is the same — the platform sells governance and identity, not the model.

• 2025-10-13 — Amazon Bedrock AgentCore goes GA:

  General availability in nine regions with VPC and PrivateLink, CloudFormation, tagging, A2A support in Runtime, MCP servers as Gateway targets, and consumption-based pricing. A cloud agent platform reaching GA with both protocols wired in is the moment the protocol layer stopped being optional.

• 2025-11-01 — Ignite 2025: Microsoft Foundry, and identity for agents:

  Microsoft renamed the platform again — Azure AI Studio to Azure AI Foundry (Ignite 2024) to Microsoft Foundry — and shipped a Responses-API-based “Agents v2” model plus Microsoft Entra Agent ID, giving agents first-class directory identities. The naming churn is noise; agents becoming principals in your identity system is not.

• 2025-12-01 — AgentCore adds Policy and Evaluations in preview:

  Policy takes natural-language rules, compiles them to Cedar, and enforces them at the Gateway — authorization outside the model, where an injection cannot argue with it. Evaluations arrived in preview with 13 built-in evaluators, alongside episodic memory and bidirectional streaming.

• 2025-12-16 — Agent Engine Sessions and Memory Bank reach GA:

  Google made conversation state and long-term memory generally available as managed services, cut Runtime pricing, added seven regions, and scheduled usage-based charging for Sessions, Memory Bank, and Code Execution from 28 January 2026. Memory became a billable dependency, which is a design constraint as much as a cost one.

• 2026-03-01 — AgentCore Evaluations, Policy, and CLI reach GA:

  Evaluations went GA in nine regions and Policy in thirteen, with the AgentCore CLI. Twelve months after the first cloud agent runtime reached GA, the parts reaching GA are measurement and authorization — the two things nobody could buy in 2023.

• 2026-05-01 — Microsoft Agent 365 reaches GA:

  Microsoft’s control plane for observing, governing, and securing agents shipped generally available, bundled into the new Microsoft 365 E7 suite. Read it with Entra Agent ID: the enterprise question moved from “can we build agents?” to “how many are running, owned by whom, with what permissions?”

• 2026-07-30 — Bedrock Agents Classic closes to new customers:

  The 2023 service became Agents Classic in maintenance mode: no new customers from this date and no new features — but AWS states there is no end-of-life date and no migration deadline for existing customers. The abstraction that launched the cloud agent era lasted under three years; plan your own portability accordingly.

• 2026-08-01 — AgentCore Agent Registry and Runtime Instances reach GA:

  The Agent Registry — a governed catalog of agents, tools, and MCP servers with AWS Organizations auto-detection and CMK encryption — went GA alongside Runtime Instances (EC2-backed sessions up to 14 days, GPU families) and AgentCore payments. AWS and Microsoft both now sell an agent inventory, which tells you what their customers were struggling with.

• 2026-08-26 — The Azure OpenAI Assistants API sunsets:

  Assistants API workloads and the azure-ai-inference package reached end of service, pushing everyone onto the GA Foundry Agents service and the openai package. Two more deadlines are already on the calendar: Foundry visual workflows retire on 1 December 2026, and Agents (classic) — the threads/runs/messages model — retires on 31 March 2027.

• 2026-09-01 — Google’s agent stack reappears as Gemini Enterprise Agent Platform:

  Observed on 13 September 2026: the Vertex AI agent docs are frozen and folded into Gemini Enterprise Agent Platform, Agent Engine is presented as Agent Runtime, and ADK has moved to adk.dev with a 2.x line (graph workflows, five languages). No announcement date for the rebrand was established, so treat this as an “observed by” marker — and as a reminder that in this space the product name is the least stable thing about it.