Salesforce’s next-generation agent development environment, “Salesforce Headless 360,” unveiled recently at its TDX event, is being described as the boldest architecture change in the company’s history.

“Headless 360” enables AI agents to directly call and execute the full range of Salesforce platform functions through APIs, MCP tools and command-line interface (CLI) commands, without people opening screens and clicking. Previously, using the Salesforce platform required people to access it through a web browser and click through screens. Headless 360 removes the screen.

A VentureBeat report said Headless 360 can be seen as Salesforce answering “no” to the question of whether a graphics UI-based CRM is still needed in an environment where AI agents can reason, plan and execute.

Zahra Eshghi Govindarajan (자예시 고빈다르잔), a Salesforce vice president, said deciding how to build an agent is only the beginning. She said all customers faced lifecycle-related issues when building agentic systems on any stack, and stressed that Headless 360 was the result of lessons learned from deploying agents to thousands of enterprise customers, not a marketing theory.

Headless 360’s features can be summarised in 3 main points.

The first is that agents can be built in any way. More than 60 new MCP (Model Context Protocol) tools and more than 30 preconfigured coding skills support external coding agents such as Claude Code, Cursor, Codex and Windsurf in accessing Salesforce’s full data, workflows and business logic. There is no need to work inside Salesforce’s own IDE. Salesforce says users can build, deploy and manage Salesforce applications through AI coding agents from any terminal.

The second feature is that it can be deployed to any screen. The Agentforce Experience Layer supports building agent functionality once while presenting a screen suited to each platform such as Slack, Teams and mobile. There is no need to build screens separately for each platform.

In a demo conducted at the Headless 360 launch, Salesforce officials deployed an experience defined once to six screens without separate code.

The third feature is building trustworthy agents. To that end, Salesforce unveiled new lifecycle management tools covering testing, evaluation, experimentation, observation and orchestration.

Salesforce also officially launched and open-sourced Agent Script, a domain-specific language that deterministically defines agent behaviour.

The company said the spread of AI agents within enterprises requires combining probability-based large language models (LLMs) with deterministic methods. It said LLM-based agents are probabilistic systems, but enterprises require deterministic outcomes, and that was why it created Agent Script. Agent Script is described as a language that combines the deterministic characteristics of programming languages with the flexibility of LLMs, allowing some steps to follow clear business logic while defining other steps for LLMs to infer freely.

At this TDX, Salesforce integrated OpenAI, Anthropic, Google Gemini, Meta Llama and Mistral AI models. But it said it remains to be seen whether MCP will become the dominant standard for AI agents. That is why it also offers APIs and CLI as options in addition to MCP.

Govindarajan said she could not be sure MCP would remain a standard. She said when it first appeared, many engineers felt it was a wrapper over a well-made CLI, and that many people still say CLI is as good as or better than MCP.

Salesforce is also shifting Headless 360 pricing from per-user licences to usage-based charging. The company said per-user pricing does not fit scenarios where agents work in place of people.