Salesforce Launches Headless 360 Agentic Platform, Redefining AI‑Driven Customer

Salesforce’s Headless 360 platform reframes the entire CRM suite as a programmable surface, converting every service—CRM, Service Cloud, Marketing Cloud, Commerce Cloud, and even Slack integrations—into an API, MCP (Model Context Protocol) server, or CLI command that can be invoked by autonomous coding agents. At the TDX developer conference, Joe Inzerillo, president of enterprise and AI technology, explained that the “experience layer” now lives outside the traditional browser UI, allowing interactions to surface in Slack, Microsoft Teams, voice assistants, ChatGPT, or custom React front‑ends (Theregister). This shift means that developers no longer need to hand‑craft UI components; instead, AI agents can generate the necessary configuration and code on demand.

The centerpiece of Headless 360 is Agentforce Code, branded as “Vibes,” a browser‑based IDE built on Visual Studio Code and bundled with the free Salesforce Developer Edition and paid tiers. According to the event briefing, Agentforce Vibes ships with pre‑configured Salesforce extensions, the Salesforce CLI, and organization metadata, streamlining the hand‑off between the IDE and the platform (Theregister). Its default large language model is Claude Sonnet 4.5, which operates in two distinct modes: “plan” for high‑level design and “act” for concrete code generation. The IDE supplies a library of agent skills that can, for example, create custom UI tabs or generate declarative Flow automations, all while targeting Apex—the Java‑like language that underpins Salesforce app development (Theregister).

Headless 360 also introduces Agent Script, a domain‑specific language for defining custom agents that can orchestrate platform actions. Salesforce announced that Agent Script will be open‑sourced, enabling third‑party developers to extend the agentic ecosystem (Theregister). However, the company cautions that agents remain probabilistic rather than deterministic, acknowledging that “agents … reason their way to unexpected outcomes” and emphasizing the need for robust testing, observability, and safety controls built into the platform’s testing center (Theregister). Usage limits for the free Developer Edition are set at 110 requests per month and 1.5 million tokens, with a refresh schedule that ends on May 31, after which a final allocation applies (Theregister).

From an architectural standpoint, Headless 360’s reliance on MCP servers signals a move toward a service‑mesh model where each Salesforce capability is exposed as a microservice endpoint. This design aligns with the broader industry trend of decoupling front‑end experiences from back‑end logic, allowing AI‑driven agents to compose workflows across disparate services without manual integration work. By treating the entire suite as a programmable API surface, Salesforce hopes to lower the barrier to entry for “anyone on the platform,” extending app‑building beyond seasoned developers to business analysts and citizen developers who can interact with the system through natural‑language prompts fed to Claude or other LLMs (Theregister).

The practical implications for enterprise customers are immediate. Companies can now instruct an AI agent to provision a new sales pipeline, attach custom fields, and expose the resulting UI in a Slack channel—all with a single natural‑language command. Because the underlying actions are executed via Apex or Agent Script, the generated artifacts remain fully version‑controlled and auditable within Salesforce’s metadata framework. This approach promises faster time‑to‑value for bespoke CRM extensions while preserving the platform’s governance and security standards. Nonetheless, the probabilistic nature of LLM‑driven code generation means that enterprises must adopt rigorous validation pipelines to catch misconfigurations before they reach production (Theregister).