For the past eighteen months, the enterprise agent conversation was mostly theoretical. AWS had early Bedrock capabilities. Google had demos. Microsoft had Copilot integrations that mostly meant “we bolted GPT-4 onto SharePoint.” None of it was production-grade infrastructure for building autonomous multi-agent systems at scale.

That changed this week.

Google announced the Gemini Enterprise Agent Platform at Cloud Next. AWS has been building out Bedrock AgentCore. Microsoft Azure AI Foundry has been steadily accumulating capabilities on top of its deep M365 integration surface. All three now have what I’d call credible enterprise offerings — not demos, not preview access, not research papers. Shipping platforms.

If you’re evaluating agent infrastructure for your team, the answer is no longer “wait and see.” You’re making a choice now, whether you intend to or not.

What Each Platform Is Actually Betting On

The marketing materials for all three platforms say roughly the same things: multi-agent orchestration, memory, tool APIs, governance. That’s the commodity layer. The differentiation is in what each company has decided is the hardest problem.

AWS Bedrock AgentCore is betting on managed tooling depth. The philosophy is: agents are complex, so we’ll abstract the failure-prone parts into managed services. Memory, knowledge bases, tool calling, orchestration flows — not primitives you wire together yourself, but managed services with SLAs. If your team is already AWS-native, the integration surface is real: ECS, Lambda, CloudWatch, IAM. The auth story is mature because it inherits from a mature cloud. The managed approach costs you control, but it buys you operational overhead you’d otherwise spend badly.

Microsoft Azure AI Foundry is betting on integration breadth. The value proposition is that your enterprise data is already in Microsoft systems — Teams, SharePoint, Exchange, Dynamics. Foundry agents can act on that data without a data pipeline replication project first. The model catalog is broad. Fine-tuning pipelines are there. If your core business runs on M365, the integration surface Microsoft offers is genuinely difficult to replicate elsewhere. The limitation is the same as it’s always been with Microsoft enterprise products: depth of integration is a feature and a lock-in at the same time.

Google Gemini Enterprise Agent Platform is betting on security-first design. The differentiators announced at Cloud Next this week — Agent Identity with cryptographic IDs, Agent Gateway for injection and poisoning defense — are not convenience features. They’re an architectural position: agents are a new security surface, and the time to address that is before deployment, not after. The ADK (Agent Development Kit) and Agent Studio follow the same logic: build structured, auditable, governable systems from the start. Google is positioning this as what regulated industries and security-conscious enterprises will eventually require. They may well be right.

The Decision Framework

The right choice depends on one question before anything else: what does your data gravity look like?

Your business runs on AWS infrastructure. Pick Bedrock AgentCore. The managed service depth is real, the IAM/VPC integration is mature, and you won’t spend six months on plumbing before your first agent delivers value. The operational overhead is lowest if you’re already AWS-native.

Your business runs on M365. Start with Azure AI Foundry. Not because it’s the best agent platform in isolation, but because the alternative — replicating your Teams, SharePoint, and Exchange data into another cloud to run agents on it — is a project that will take longer than building the agents. Gravity wins.

Security and auditability are your primary constraint. Look harder at Google. The Agent Identity and Agent Gateway story is early, but it’s the only platform that shipped with a named answer to the agent security question. If you’re in financial services, healthcare, or any regulated context where you need to explain what your agents did and prove they weren’t compromised, Google’s architectural bet is worth taking seriously — even if it means getting ahead of your current cloud strategy.

You’re genuinely cloud-agnostic. This is rare, but if it’s true: Google’s security story plus AWS’s managed depth are a stronger combination than either alone. Build agents on Bedrock, enforce identity and gateway policies through Google’s framework as it matures, avoid the Microsoft lock-in unless your M365 integration surface justifies it.

What This Week Changed

The agent platform race is over in one sense: all three companies now have offerings serious enough to evaluate. The question from here isn’t who shipped — it’s who you’re betting your agent architecture on for the next three years.

That’s an architectural decision, not a procurement decision. It compounds. Make it intentionally.