Strip away the product names, and the message is this: the vendor that spent 25 years making a browser-based CRM the center of the sales workflow just rebuilt itself so agents never have to touch the browser. Salesforce did not do this because they wanted to. They did it because agents are the new users, and agents do not click. They call APIs, invoke tools, and execute commands.

The software sector has lost roughly 28% since September. The thesis behind the sell-off is simple: if agents can reason, plan, and act, traditional SaaS interfaces become a tax rather than a feature. Headless 360 is a direct answer to that thesis. For Salesforce, headless is survival.

Every enterprise has to do the same math. Systems that can only be used through a UI are invisible to agents. Anything invisible to agents is about to become a bottleneck.

The announcement glosses over the hard part. Exposing APIs is the easy piece. Making them usable by agents, at scale, across the enterprise, is a different problem.

Agents need more than endpoints. They need reliable discovery, so they can find the right tool instead of guessing. They need context that is fresh, scoped, and trustworthy. They need governance that enforces authN, authZ, rate limits, and data boundaries at runtime. They need cost controls, because an agent in a loop can burn budget faster than any human. And they need support for the protocols agents actually speak, which now includes MCP and A2A alongside REST and events.

Any enterprise that "goes headless" without solving these pieces ends up with a broader attack surface, not a working agent platform. A thousand APIs with no discovery or governance layer is not agent-ready infrastructure. It's a liability.

Salesforce also introduced Agent Fabric, a governance plane for agents operating inside the Salesforce platform. Fine. Every major vendor will ship their own version. ServiceNow will. Workday will. SAP will. Databricks already is.

Each one will be excellent at governing agents inside its own platform. None of them will govern agents across your platforms. And that is where the actual enterprise lives. And this doesn’t even start to consider the risks and opportunities related to *external *agent access control, governance, and [_monetization_](https://konghq.com/blog/enterprise/the-context-economy)_monetization_. 

Your agents will not operate inside just one vendor's walls. They will pull data from Salesforce, inventory from SAP, tickets from ServiceNow, telemetry from Splunk, and domain APIs your team built last quarter. The governance problem is not "how do I control agents inside Salesforce?" — it's "how do I control agents across every system they touch?"

If the answer is one fabric per vendor, the problem isn't solved. The silos just moved up a layer. That is the [AI fragmentation tax](https://konghq.com/blog/enterprise/hidden-ai-fragmentation-tax)AI fragmentation tax playing out in real time.

The architecture that services agents is more than APIs. It's a connectivity layer that handles the full data path agents rely on. Synchronous APIs. Asynchronous events. AI-native protocols like MCP and A2A. Under a single control tower that enforces policies, logs every call, meters cost, and routes across environments.

This is not a wish list. Every item on it exists because agents have already broken something in production without it. Unauthorized tool calls. Runaway token spend. Context leakage. Ungoverned MCP servers installed by developers in a weekend sprint.

The organizations that handle this well over the next 18 months will share a few traits. They will treat API-first as a default, not a project. They will treat events and AI protocols as first-class alongside REST. And they will have a single governance plane across every interface an agent can reach, regardless of which vendor owns the system behind it.

We’ve been thinking about this for a while. In fact, [_we called this earlier in the year_](https://www.linkedin.com/posts/sonicaghi_software-is-becoming-headless-activity-7429404583912792065-c8qJ?utm_source=share&utm_medium=member_desktop&rcm=ACoAACYKmYsBZxivSum9b2npyzxl_5MWRq_nRhc)_we called this earlier in the year_. And, we’ve been [_singing the “No AI without API” song_](https://www.linkedin.com/posts/markwoneill_all-the-webs-an-api-and-ai-agents-its-activity-7450886846903087104-zIo0?utm_source=share&utm_medium=member_desktop&rcm=ACoAACYKmYsBZxivSum9b2npyzxl_5MWRq_nRhc)_singing the “No AI without API” song_ for even longer.

From a product perspective, Kong has spent the last decade building the connectivity primitives this architecture requires. API Gateway, Event Gateway, AI Gateway, MCP Registry, and Agent Gateway aren't separate products so much as one control plane for every protocol an agent speaks.

[Context Mesh](https://konghq.com/blog/product-releases/introducing-kong-context-mesh)Context Mesh extends that control plane to the data side, so agents get scoped, governed context instead of just raw database access or APIs that aren’t “agent-ready.” This will be a major difference between “just headless” and “agent-native headless.”

The point is not that Kong is the only way to solve this. The point is that Salesforce just validated, at scale, that the era of UI-centric enterprise software is ending. What replaces it has to be built on connectivity that is open, governed, and multi-vendor by default. A single vendor's fabric, however well engineered, cannot govern an enterprise that runs on many vendors.

Headless 360 is a strong move for Salesforce. It's also a warning for everyone else. If your architecture still assumes a human will log in and click, you're already behind. The agents are already not clicking.