The Interface Is No Longer the Contract

The Interface Is No Longer the Contract

For most of the history of enterprise software, the contract between a vendor and a customer was the user interface. You licensed seats, people logged in, and the value showed up in the screens. Integration was an afterthought, bolted on through APIs that only engineers touched.

That contract is breaking. Three things changed at once.

Assistants became where people work

Knowledge workers now start a large share of their tasks in a general-purpose assistant rather than in a specific application. They ask a question, draft a document, analyze a spreadsheet, and increasingly expect the assistant to reach into the systems that hold the answer. The interface people want is a conversation, and they want to choose which one.

A common protocol arrived

The Model Context Protocol (MCP) gave assistants a standard way to connect to systems. Introduced in late 2024, it was adopted by OpenAI in early 2025 and by Google shortly after, gained secure remote sign-in later that year, and was donated to the Linux Foundation at the end of 2025. Within roughly eighteen months, opening a system to every assistant went from a custom integration project to a supported pattern. A companion protocol, Agent2Agent (A2A), does the same for agents talking to other agents.

Every vendor reorganized around agents

The large platforms have all moved. Salesforce, Microsoft, ServiceNow, SAP, Oracle, Workday and Atlassian now each ship agents, expose their platforms to outside AI through MCP, and are rethinking how they price. Some of their leaders have gone further and argued that the application itself will collapse into agents — that business logic will move out of the screens and into the AI that operates on the data.

What "headless" actually means

Headless doesn't mean no interface. It means the interface stops being the product. A sensible, widely shared counter-view is that applications become review surfaces: places where people check, approve and correct work that agents prepared. We think that's right for most of the enterprise for most of the next decade.

So the headless enterprise isn't one where people stop using screens. It's one where:

• People reach the work from the interface they choose — an assistant, a chat tool, an app.
• AI does a growing share of the preparation, assembly and follow-up.
• The systems underneath are judged by what they let people and AI do, not by how good their screens are.
• Decisions still belong to people, and the record of those decisions has to live somewhere every interface can see.

That last point is where most enterprises aren't ready.