Forget individual AI assistants. The next wave of enterprise automation involves teams of AI agents — one that handles customer queries, another that checks inventory, a third that updates your CRM — all working together without human supervision. But here’s the problem no one has solved: how do these agents coordinate?
Research into “foundation protocols” for agentic systems is gaining momentum, and it points to an emerging battleground. The company that controls how AI agents discover, authenticate, and transact with each other will effectively control the marketplace for business process automation. If this sounds like the browser wars or the cloud platform wars, that’s because it is.
Why Coordination Is the New Bottleneck
Most enterprises experimenting with AI agents today are hitting the same wall. A single agent connected to a large language model can do impressive things — summarise documents, draft emails, answer questions. But production workflows require multiple agents to hand off tasks, share context, and trust each other’s outputs.
Right now, that coordination is bespoke. Engineering teams stitch together agents using custom code, and every new integration means more technical debt. Identity management — knowing which agent is authorised to do what — is a mess. Incentive structures, like how to bill when one vendor’s agent uses another’s, barely exist.
These aren’t theoretical concerns. They are the reasons multi-agent deployments stall after pilot phase. Research proposing standardised coordination layers is an attempt to clear this bottleneck, but standards don’t emerge in a vacuum. Someone builds them, and that someone usually captures value downstream.
Microsoft and OpenAI Are Already Positioning
OpenAI’s agent-building tools and Microsoft’s Copilot ecosystem are not just products — they are platform plays. Microsoft has been embedding agent orchestration capabilities into Azure, making it easy to spin up agents that work within its cloud environment. OpenAI’s function-calling features and plugin architecture hint at similar ambitions.
The pattern is familiar: offer convenience now, create switching costs later. If your agents are built on Azure’s coordination layer, moving them to AWS or Google Cloud becomes a migration project, not a configuration change. If your automation stack depends on OpenAI’s agent protocols, you’re betting on OpenAI’s roadmap.
Neither company has declared an intent to own “the” coordination standard. But both are building the infrastructure that could become one by default. In enterprise software, de facto standards often beat formal ones simply by being there first.
New Procurement Headaches Are Coming
For CIOs, this trend creates immediate planning questions. Multi-agent automation is not a single purchase; it is an architecture decision. Vendors are beginning to offer agent marketplaces — curated collections of pre-built agents that work within their orchestration layer. Choosing a marketplace means choosing a governance model, an identity framework, and a commercial relationship you may not fully control.
Expect new procurement categories to emerge: orchestration-as-a-service, agent identity management, inter-agent billing platforms. Each will come with its own vendor lock-in risks. The CIO who signs up for a convenient all-in-one solution in 2025 may find themselves locked into that vendor’s ecosystem by 2027.
Governance is the other headache. When agents from different vendors interact, who is liable when something goes wrong? Current contracts are not written for this. Legal and compliance teams will need to catch up fast.
Founders Face a Compatibility Gamble
If you’re building an enterprise automation product, the coordination layer question is strategic. Design your agents to work only within your own stack, and you control the experience but limit distribution. Design for compatibility with emerging protocols, and you gain access to partner networks but depend on standards you don’t control.
The worst outcome is building for a protocol that loses. Founders should watch which coordination approaches gain traction among large system integrators and enterprise buyers. Fast followers often do better than early adopters when standards are still fluid.
What This Means for You
The immediate action is to audit your current AI agent experiments. Ask your teams: how are these agents coordinating today, and what happens when we add a third or fourth agent from a different vendor? If the answer involves significant custom code, you are accumulating technical debt that will cost more to unwind later.
For procurement, start asking vendors about their orchestration roadmap. Do they plan to support open coordination protocols, or are they building proprietary layers? The answer will tell you how much flexibility you’ll have in two years.
For founders, design modularity into your agent architecture now. Make it possible to swap coordination layers without rewriting core logic. The standard that wins may not be the one you’re building on today.
The real story here is not about any single protocol. It is about control. Whoever owns the coordination layer owns the toll booth for enterprise automation. That’s worth paying attention to before you commit.