In Part 1, we looked at how LinkedIn and Amazon are each building structured knowledge infrastructure for AI agents — from LinkedIn's CAPT playbooks to Amazon's four-layer stack of Knowledge Bases, AgentCore Policy, Evaluations, and open-source Agent SOPs. Now let's look at how Google, Microsoft, and Salesforce are approaching the same problem, and what the pattern reveals.

Google: The Agent Marketplace With Built-In Governance

Google's Vertex AI Agent Builder is approaching the problem through what might be the most enterprise-familiar lens: centralized tool governance with a marketplace model.

Their Enhanced Tool Governance, announced in late 2025, integrates the Cloud API Registry directly into Agent Builder. Administrators can now manage a centralized catalog of tools that agents are allowed to use — not just at the model level, but at the organizational level. Every API exposed through Apigee can be transformed into a governed MCP server. The tools are curated, the access is controlled, and the catalog is centralized.

But the most interesting move is Gemini Enterprise — an internal agent marketplace where organizations can publish custom AI agents with controlled sharing, centralized governance, and monitoring. Think of it as an enterprise app store for agents, where every agent gets an identity tied to Cloud IAM, policies are enforced through Model Armor, and the organization maintains a single access point for discovery and oversight.

Agents get their own native identities within Google Cloud. You can enforce true least-privilege access, establish granular policies, and set resource boundaries to meet compliance requirements. The observation layer tracks token usage, latency, and error rates, while an evaluation layer simulates user interactions to test agent reliability before deployment.

Google's bet: if you give the enterprise familiar patterns — identity management, access control, marketplaces, monitoring dashboards — they'll adopt agent governance the same way they adopted cloud governance.

Microsoft: Agent Identities and the Compliance Stack

Microsoft's approach with Copilot Studio is perhaps the most governance-forward of all, which makes sense given that their customers are often the most compliance-conscious enterprises on the planet.

In 2025, Copilot Studio introduced automatic creation of Microsoft Entra agent identities. Every agent gets provisioned with an identity, enabling IT teams to apply policies, monitor behavior, and secure access with enterprise-grade oversight. This is significant because it treats agents not as tools but as entities — the same way you'd provision a new employee with credentials, roles, and access boundaries.

Human-in-the-loop controls, now in preview, allow organizations to require human review or approval at specific stages of an agent's execution. Not for every action, but for sensitive data interactions, downstream business triggers, and decisions that require accountability.

The integration with Microsoft Purview and Sentinel for end-user activity auditing means that agent actions are logged, searchable, and subject to the same compliance workflows as human employee activities. Tenant-wide inventory in the Power Platform admin center gives administrators visibility into which agents exist, what they're doing, and how they're performing across every environment.

Microsoft's thesis: agents should fit into the governance infrastructure enterprises already have, not require a new one. Identity, compliance, audit, and access control — extended from people to agents.

Salesforce: "Governance Will Shift From Optional to Mandatory"

Salesforce is saying the quiet part out loud.

Their Agentforce platform embeds what they call the Einstein Trust Layer directly into agent architecture — real-time grounding, data security controls, and accuracy validation as core infrastructure rather than optional add-ons. With Agentforce 360, launched in late 2025, the platform embeds three pillars of trust — data quality, security, and employee adoption — directly into how agents reason, act, and collaborate with humans.

Inspector agents provide always-on analytics: why an agent acted, what data influenced it, and whether the outcome was successful. This isn't debugging. It's the audit trail that compliance, risk, and ROI measurement all depend on.

But the sharpest insight from Salesforce's ecosystem is this: "Getting Agentforce buy-in at organizations starts with a clear governance story." Many administrators can't get agent adoption approved without one. Governance isn't slowing down deployment — the absence of governance is. The companies that can't explain how agents are controlled are the ones stuck in pilot.

Salesforce's leaders predict that by 2026, "governance will shift from optional to mandatory as agents gain the ability to trigger real business actions." Trust frameworks, observability, and deterministic guardrails will be embedded directly into agent architectures, not bolted on after deployment.

The Pattern No One Is Naming

Step back, and the convergence is striking.

LinkedIn is building organizational context delivery. Amazon is building deterministic policy enforcement. Google is building governed tool marketplaces. Microsoft is building agent identity and compliance. Salesforce is building trust layers and audit infrastructure.

Five companies, five approaches, and yet every single one arrived at the same conclusion independently: agents need organizational knowledge that is structured, governed, and enforced — not just retrieved.

Zapier's 2026 State of Agentic AI survey adds the demand side to the picture. Among 200 senior technology leaders at enterprises with 5,000+ employees: 25% expect to reach full-scale agent orchestration this year, 43% anticipate agentic AI models operating with limited human intervention, and — critically — 71% identified "human-in-the-loop approvals" as their top governance priority for 2026. Not better models. Not faster inference. Governance.

And yet, despite this convergence, there's a gap in the middle of the picture.

Each of these platforms is solving their piece of the governance puzzle, within their ecosystem, for their type of agent, on their infrastructure. Amazon governs agents on Bedrock. Microsoft governs agents in Copilot Studio. Salesforce governs agents in Agentforce. Google governs agents in Vertex.

But the enterprise reality — as anyone deploying agents at scale already knows — is that your organization doesn't run on one platform. Your marketing team uses one tool. Sales uses another. Support uses a third. Customer success built something custom. The governance challenge isn't governing agents within a single platform. It's ensuring that every agent, across every platform, operates from the same organizational knowledge: the same pricing rules, the same brand voice, the same compliance boundaries, the same approved messaging.

No single platform solves that today.

What This Tells Us About What's Coming

The fact that Amazon, LinkedIn, Google, Microsoft, and Salesforce are all investing in agent knowledge infrastructure at the same time isn't a coincidence. It reflects a maturation in how the industry understands the agentic AI challenge.

Phase 1 was model capability: make agents smarter. That problem is largely solved — or at least, it's advancing faster than organizations can absorb.

Phase 2, happening right now, is organizational alignment: make agents know how your company works. This is the knowledge base problem, and the approaches above show real progress.

Phase 3, the one most enterprises haven't reached yet, is organizational control: make agents accountable. Not just informed, but governed. Not just smart, but constrained. Not just capable, but trustworthy.

LinkedIn nailed the diagnosis: the bottleneck isn't model sophistication, it's organizational context. Amazon nailed the enforcement: policy must be deterministic, not probabilistic. Salesforce nailed the business case: without governance, you can't even get to deployment.

The companies that put these pieces together — organizational knowledge, deterministic enforcement, cross-platform consistency, audit trails, and role-based ownership — won't just have better agents. They'll be the ones whose agents are allowed to do more, because the organization trusts them to do it right.

Deloitte found that only 21% of companies deploying agentic AI have a mature governance model. The other 79% are watching the same convergence described above and trying to figure out which platform will solve the problem for them. The emerging answer is: none of them will, alone. The governance layer that enterprises need is one that works across all of them.

The race to build agentic knowledge bases has already started. Amazon, LinkedIn, Google, Microsoft, and Salesforce are all running. The question for every other organization is whether you're building the governance layer that ties it all together — or hoping one of them will build it for you.

References

  • LinkedIn Engineering — "Contextual Agent Playbooks and Tools: How LinkedIn Gave AI Coding Agents Organizational Context" (Jan 2026): https://www.linkedin.com/blog/engineering/ai/contextual-agent-playbooks-and-tools-how-linkedin-gave-ai-coding-agents-organizational-context
  • The New Stack — "AWS' New Policy Layer in Bedrock AgentCore Makes Sure AI Agents Can't Give Away the Store" (Dec 2025): https://thenewstack.io/aws-new-policy-layer-in-bedrock-agentcore-makes-sure-ai-agents-cant-give-away-the-store/
  • AWS — "Amazon Bedrock AgentCore Adds Quality Evaluations and Policy Controls for Deploying Trusted AI Agents" (Dec 2025): https://aws.amazon.com/blogs/aws/amazon-bedrock-agentcore-adds-quality-evaluations-and-policy-controls-for-deploying-trusted-ai-agents/
  • AWS Open Source Blog — "Introducing Strands Agent SOPs — Natural Language Workflows for AI Agents" (2026): https://aws.amazon.com/blogs/opensource/introducing-strands-agent-sops-natural-language-workflows-for-ai-agents/
  • Google Cloud Blog — "New Enhanced Tool Governance in Vertex AI Agent Builder" (2025): https://cloud.google.com/blog/products/ai-machine-learning/new-enhanced-tool-governance-in-vertex-ai-agent-builder
  • Microsoft — "6 Core Capabilities to Scale Agent Adoption in 2026" (2026): https://www.microsoft.com/en-us/microsoft-copilot-studio/blog/copilot-studio/6-core-capabilities-to-scale-agent-adoption-in-2026/
  • Salesforce — "Welcome to the Agentic Enterprise: Agentforce 360" (Oct 2025): https://www.salesforce.com/news/press-releases/2025/10/13/agentic-enterprise-announcement/
  • Zapier — "State of Agentic AI Adoption Survey" (2026): https://zapier.com/blog/ai-agents-survey/
  • Deloitte — "The State of AI in the Enterprise, 2026 AI Report": https://www.deloitte.com/global/en/issues/generative-ai/state-of-ai-in-enterprise.html