PulseFlow Tecnologia
← Insights
·6 min read·market-analysis·microsoft-foundry·gemini-enterprise·aws-bedrock
PulseFlow Tecnologia

Microsoft, Google, and AWS: three strategies for enterprise agents

The world's three largest clouds arrived at the same problem - how to sell AI agents to enterprises - through three completely different entry points. Comparing each company's recent announcements side by side makes clear this isn't just marketing: these are distinct architectural bets, built on top of what each company already dominated before the agent era.

Microsoft: building out from the productivity and development ecosystem

At Build 2026, Microsoft reinforced Microsoft Foundry as the central platform for building and running agents at scale, complemented by two specific launches: Foundry IQ, with "unified knowledge and serverless retrieval" and gains on agentic RAG benchmarks; and Work IQ, "production-ready intelligence for every agent," with its own APIs to integrate agents into existing workflows. On the governance side, the Agent Control Specification promises "portable runtime governance" - a control policy that isn't locked to a single platform. Even the infrastructure was redesigned for this: Azure Cobalt 200 VMs deliver a 50% performance gain, "fully optimized for modern agentic AI workloads." Microsoft's bet is clear: use its already-installed base of Microsoft 365, GitHub, and Azure as the distribution surface for agents.

Google: rebuilds the entire AI platform around agents

Google Cloud didn't treat agents as a new product inside Vertex AI - it reorganized the entire Vertex AI within the Gemini Enterprise Agent Platform, around four capabilities (build, scale, govern, optimize) and access to more than 200 models, including competitor Anthropic's. This is a full-platform bet, not an incremental one: Google's entire AI services roadmap now gets delivered through this agent layer, no longer as standalone services. Cases like Payhawk (over 50% reduction in expense-report time) and more than 6 trillion tokens processed monthly via the Agent Development Kit show the scale this already operates at.

AWS: attacks data and execution infrastructure, not the user experience

While Microsoft and Google build around productivity and the model, AWS focuses on the infrastructure problem behind the agent: Managed Knowledge Base solves corporate data connectivity (six native connectors, smart parsing, multihop retrieval) without every team building its own RAG pipeline. It's a quieter bet, closer to engineering - AWS isn't trying to be the interface the end user sees, it's trying to be the layer that any agent, from any model provider, uses to access corporate data securely.

What this means for platform buyers

None of the three strategies is objectively "better" - they respond to different starting points. A company heavily dependent on the Microsoft 365 and GitHub ecosystem has a shorter path with Foundry. A company already running most of its workload on Vertex AI benefits from the reorganization around the Gemini Enterprise Agent Platform. And a company with fragmented data architecture, needing a neutral connectivity layer, finds in AWS a partner designed not to impose a specific user experience. The right decision depends less on which company has "the best agent" and more on which of these three architectural bases already underpins the company's operation today.

Sources