The Reality
Salesforce Agentforce for sales and service. Microsoft Copilot for employee productivity. HubSpot Breeze for marketing/inbound. ServiceNow for ITSM. Each vendor’s agent optimal for their domain.
Gartner’s late-2025 enterprise AI survey found 78% of large organizations run agents from three or more vendors. Each vendor has a structural advantage in its native domain because the agent has tightest data access there: Agentforce sees Sales Cloud opportunities natively; Copilot sees Outlook and Teams; Breeze sees marketing engagement; Now Assist sees ITSM tickets. Forcing one vendor across all domains usually means weaker integration, higher latency, and more brittle workflows than embracing the multi-vendor reality.
Coordination Layer
Salesforce Agent Fabric positions as the neutral orchestration layer. Microsoft has its own orchestration. Specialized orchestrators (LangGraph, CrewAI) emerge. Pick one; don’t try to orchestrate in N places.
Three orchestration patterns. Vendor-led: one platform’s orchestrator (Agent Fabric, Microsoft Copilot Connectors) acts as the front door, delegating to others via MCP or REST. Neutral: an open framework (LangGraph, CrewAI) routes requests to whichever vendor agent fits, with the framework owning state and audit. Federated: each vendor agent stays in its lane, with handoffs through shared state in a CDP. Most enterprises end up with a hybrid: vendor-led inside each domain, neutral orchestration for cross-domain workflows.
Data and Identity
Unified identity and data become prerequisite. Data 360 or equivalent CDP as foundation. Agents from different vendors must see consistent customer data or they contradict each other in user-facing interactions.
Required foundations. Single customer ID resolved across CRM, marketing, support, and product systems — typically via Data Cloud, Treasure Data, or Hightouch with reverse ETL. Single sign-on through Okta or Entra ID with claims that propagate to each agent. Shared consent and preference management so a customer’s “no marketing” preference reaches Breeze and Agentforce identically. Without this, multi-vendor agents will give the customer contradictory answers within an hour.
Governance
Policy consistency across vendors. Cost allocation across vendors. Incident response across vendors. Multi-vendor AI isn’t free — it’s additional coordination cost. Budget accordingly.
Coordination overhead categories. Policy translation: each vendor expresses guardrails differently (Trust Layer, Copilot Studio policies, Now Assist Skills config) — maintain one policy spec and translate per vendor. Cost rollup: each vendor bills separately; build a unified dashboard in Vantage, Apptio, or a custom Snowflake mart. Incident response: a customer-facing failure may originate from any vendor — a unified incident commander role and shared status page are non-negotiable. Procurement: triple the contract complexity, especially for AI Act addenda.
Common Failure Modes
Five recurring patterns. Two vendor agents giving contradictory answers because their data sources diverged. Per-vendor cost ceilings that sum to 3x the actual budget because no rollup exists. Customer escalation paths that loop between vendors with no human owner. Inconsistent disclosure language (“you’re chatting with Einstein” vs “with Copilot”) confusing customers. Per-vendor evaluation suites that don’t share a regression baseline.
What to Do This Week
Map every customer-facing AI agent in production by vendor and tag the data source it reads from — surface any contradictions before customers do.
