Salesforce on April 15, 2026, unveiled what it described as the most ambitious architectural transformation in its 27-year history: a sweeping initiative called Headless 360 that exposes every capability in its platform as an API, Model Context Protocol (MCP) tool, or CLI command - removing the browser as the mandatory interface for interacting with the system.

The announcement, made at the company's annual TDX developer conference in San Francisco, shipped more than 100 new tools and skills immediately available to developers. Salesforce Co-Founder Parker Harris framed the underlying logic bluntly in a question posed the previous month: "Why should you ever log into Salesforce again?"

The answer embedded in Headless 360 is that, in an enterprise environment where AI agents are doing increasing amounts of work alongside humans, the assumption that work must happen through a graphical interface no longer holds.

The three-pillar architecture

Salesforce Headless 360 is organized around three distinct pillars. Understanding the technical specifics of each matters because they address genuinely different problems in how enterprises deploy and manage AI agents.

The first pillar - build any way you want - delivers more than 60 new MCP tools and 30-plus preconfigured coding skills. These give external coding agents like Claude Code, Cursor, Codex, and Windsurf complete, live access to a customer's entire Salesforce organization, including data, workflows, and business logic. Developers no longer need to work inside Salesforce's own IDE. They can direct AI coding agents from any terminal to build, deploy, and manage Salesforce applications.

Agentforce Vibes 2.0, Salesforce's native development environment, accompanies this with what it calls an "open agent harness" supporting both the Anthropic agent SDK and the OpenAI agents SDK. Developers can choose between Claude Code and OpenAI agents depending on the task, with the harness dynamically adjusting available capabilities based on the selected agent. The environment adds multi-model support including Claude Sonnet and GPT-5, along with full organizational awareness from the start. A DevOps Center MCP brings that same programmatic access into CI/CD pipelines, enabling natural language deployment descriptions that the agent then executes - cutting cycle times by as much as 40%, according to Salesforce.

Native React support represents a significant technical addition. During the TDX keynote, presenters built a fully functional partner service application using React - not Salesforce's own Lightning framework - that connected to organizational metadata via GraphQL while inheriting all platform security primitives. This opens substantially more expressive front-end possibilities for developers who want full control over the visual layer without sacrificing the security and data access the platform provides.

The second pillar - deploy on any surface - centers on a new Agentforce Experience Layer. The core design choice here is separating what an agent does from how it appears. The layer renders rich interactive components - flight status cards, rebooking workflows, decision tiles, data layouts - natively across Slack, mobile apps, Microsoft Teams, ChatGPT, Claude, Gemini, and any client that supports MCP apps. According to Salesforce, presenters at the keynote defined an experience once and deployed it across six different surfaces without writing surface-specific code. Build once, render everywhere is the stated objective.

Custom AI agents on Slack have grown 300% since January, according to Salesforce, illustrating why the conversation-as-interface framing matters practically. Slackbot is positioned in the announcement as the front door to what Salesforce calls the Agentic Enterprise. Work that previously required navigating a separate application - approvals, decisions, workflow completions - can now surface inside the channels where teams already operate.

The third pillar - build agents you can trust at scale - addresses a problem that has emerged consistently as enterprises move AI agents from testing to production. Agents are probabilistic systems; they do not behave identically across every interaction. That fundamental characteristic has made post-launch maintenance painful. Salesforce describes customers getting agents into production through considerable effort, then discovering the system was brittle: a single change could cascade unpredictably through the agent's behavior, requiring all prior testing to be redone.

Agent Script and the determinism problem

The response to that brittleness is Agent Script, a domain-specific language for defining agent behavior deterministically, which Salesforce is making generally available and open-sourcing with this announcement. Jayesh Govindarjan, EVP of Salesforce and one of the architects behind Headless 360, described it to VentureBeat as a programming language that "brings together the determinism that's in programming languages with the inherent flexibility in probabilistic systems that LLMs provide."

The language functions as a single flat file - versionable, auditable - that defines a state machine governing how an agent behaves. Within that machine, enterprises specify which steps must follow explicit business logic and which can reason freely using LLM capabilities. Salesforce open-sourced Agent Script at TDX, and Govindarjan noted that Claude Code can already generate it natively because of the quality of its documentation.

The third pillar also introduces a Testing Center that surfaces logic gaps and policy violations before deployment, Custom Scoring Evals that let enterprises define what a correct agent output looks like for their specific use case, and an A/B Testing API that enables running multiple agent versions against real traffic simultaneously. After launch, Observability and Session Tracing show not just what happened but the reasoning behind it. When an agent drifts from expected behavior, the tooling is designed to locate the cause in hours rather than weeks.

For enterprises running agents across multiple platforms and vendors, a new Agent Fabric brings them under one governed control plane, with deterministic orchestration and centralized governance across LLM, tool, and agent configurations.

Two architectures for two contexts

Govindarjan drew a distinction between two fundamentally different agentic architectures that Salesforce says every enterprise will eventually need to maintain.

Customer-facing agents - those deployed to interact with end customers for sales or service - demand tight deterministic control. Agent Script encodes these as a static graph: a defined sequence of steps with LLM reasoning embedded within each one. The structure is fixed; the reasoning within each step is flexible.

The contrasting architecture is what Govindarjan called the "Ralph Wiggum loop" - a dynamic graph that unrolls at runtime, where the agent autonomously decides its next step based on what it learned in the previous step, discarding dead-end paths and spawning new ones until the task is complete. This architecture is suited to employee-facing scenarios: developers using coding agents, salespeople running deep research, marketers generating campaign materials - contexts where an expert human reviews the output before it is used.

The critical technical detail is that both architectures run on the same underlying platform and the same graph engine. "This is a dynamic graph. This is a static graph. It's all a graph underneath," Govindarjan told VentureBeat. A unified runtime spanning tightly controlled customer interactions and free-form autonomous loops means enterprises do not need to maintain separate platforms for different agent types.

Pricing model shift

Underpinning the platform announcement is a change in how Salesforce charges for Agentforce. The company is moving from per-seat licensing to consumption-based pricing. Govindarjan described the shift as "a business model change and innovation for us." When agents rather than humans are doing the work, charging per user no longer maps to value delivered. The consumption model attempts to align pricing with the actual volume of work agents perform.

The MCP question and protocol hedging

One of the more candid moments in Salesforce's TDX communications came from Govindarjan's assessment of MCP itself - the protocol Anthropic created that has become a de facto standard for agent-tool communication. "To be very honest, not at all sure" that MCP will remain the standard, he said, according to VentureBeat. "When MCP first came along as a protocol, a lot of us engineers felt that it was a wrapper on top of a really well-written CLI."

The "Headless 360" naming reflects this uncertainty directly. Rather than betting on a single protocol, Salesforce exposes every capability across all three access patterns - API, CLI, and MCP - insulating the platform against protocol shifts. The approach is pragmatic rather than ideological: offer all three and let adoption patterns determine which gains dominance.

This hedging is relevant context for the marketing and ad tech community, which has watched MCP adoption accelerate across advertising platforms throughout 2025. Google released an open-source MCP server for its Ads API on October 7, 2025, enabling AI tools to query advertising campaigns through natural language. AppsFlyer launched its own MCP tool on July 17, 2025, targeting mobile marketing measurement. Google Analytics followed on July 22, 2025 with an experimental open-source MCP server for natural language data queries.

The question of which protocols endure matters practically. IAB Tech Lab published a comprehensive agentic AI roadmap on January 6, 2026, designed specifically to prevent the fragmentation that results when incompatible protocols proliferate. Salesforce's three-protocol strategy is one answer to that problem; industry standardization is another.

The AgentExchange marketplace and the builders fund

AgentExchange brings together 10,000 Salesforce apps, 2,600-plus Slack apps, and 1,000-plus Agentforce agents, tools, and MCP servers from partners including Google, Docusign, and Notion - all discoverable through AI-guided search and activated in one click. Salesforce provided several concrete partnership numbers: Notion cut its average sales cycle from four months to three weeks after listing on AgentExchange. Docusign processed more than 200 private offers in Q4 2025 with 60% faster time to signature. MeshMesh closed its first Fortune 500 customer six weeks after listing.

A new $50 million Builders Fund provides investment, engineering support, and go-to-market pathways for what Salesforce calls Agentblazers - developers and partners building on the Agentforce platform.

Context from a production deployment

Engine, a B2B travel management company, appeared prominently in the TDX keynote as a production case. According to Salesforce, Engine built its customer service agent, Ava, in 12 days using Agentforce and now handles 50% of customer cases autonomously. The company runs five agents across customer-facing and employee-facing functions, with Data 360 at the center of its data infrastructure and Slack as its primary workspace.

Elia Wallen, CEO of Engine, is quoted in the Salesforce announcement: "With Agentforce, we've been able to deploy sophisticated, production-ready AI agents in just 12 days, driving millions in savings while significantly increasing our technical velocity."

Why the marketing community should pay attention

The Headless 360 announcement lands at a specific moment in the agentic AI build-out. Research published in March 2026 based on a survey of 1,026 developers and technology leaders found that nearly 80% of respondents were already directly involved in building or influencing agentic AI decisions at their organizations. Full autonomy remains rare - only 4% of teams in that survey allow agents to act without any human approval. The majority operate in configurations where agents handle some actions automatically and route others for human review.

Salesforce's Headless 360 is designed for precisely this moment: production deployments that need governance and testing infrastructure, not just connectivity. The combination of Agent Script's determinism controls, the Testing Center's pre-launch verification, and the A/B Testing API's post-launch experimentation tools addresses the brittleness problem that has slowed enterprise agent deployment in practice.

For marketing technology professionals specifically, the platform is increasingly relevant infrastructure. Salesforce has organized its platform into four layers: Data 360 as the system of context, Customer 360 apps as the system of work, Agentforce as the system of agency, and Slack as the system of engagement. Headless 360 opens every layer via programmable endpoints. An AI coding agent that previously needed to navigate Salesforce's UI to access customer context - open escalations, renewal dates, SLA breaches - can now call that data directly through an API or MCP tool.

UK regulators published a detailed cross-regulatory assessment of agentic AI systems on March 31, 2026, produced jointly by the CMA, FCA, ICO, and Ofcom, noting that governance requirements are accumulating faster than many deployments have anticipated. Salesforce's choice to open-source Agent Script and build auditable, versionable state machines into its agent framework positions the platform explicitly within those governance requirements - providing the auditability and traceable logs that regulatory frameworks are beginning to demand.

The iShares Expanded Tech-Software Sector ETF had fallen roughly 28% from its September 2025 peak by the time of the TDX announcement, according to VentureBeat, reflecting investor concern that AI could erode traditional SaaS revenue models. Headless 360 is Salesforce's structural answer to that concern: by making its platform the most agent-accessible substrate available, the company attempts to turn AI from a threat into a distribution mechanism.

Timeline

Summary

Who: Salesforce, a 27-year-old enterprise software platform serving organizations worldwide, announced Headless 360 at its annual TDX developer conference in San Francisco. Key figures include Salesforce Co-Founder Parker Harris and Jayesh Govindarjan, EVP, who has been the lead architect of the initiative. Production customer Engine, a B2B travel management company, was featured as a case study.

What: Headless 360 is a platform-wide architectural initiative that exposes every Salesforce capability as an API, MCP tool, or CLI command. The announcement shipped more than 100 new tools and skills, introduced the Agentforce Experience Layer for cross-surface agent deployment, open-sourced Agent Script, launched a Testing Center and Custom Scoring Evals for pre-launch validation, added A/B Testing and Observability tooling for post-launch governance, and introduced a new Agent Fabric for multi-vendor orchestration. AgentExchange was expanded to 10,000 Salesforce apps, 2,600-plus Slack apps, and 1,000-plus Agentforce agents and MCP servers. A $50 million Builders Fund was announced alongside the initiative. Pricing for Agentforce is transitioning from per-seat to consumption-based.

When: The announcement was made on April 15, 2026, at the TDX developer conference. Salesforce states all 100-plus tools and skills are immediately available to developers as of that date.

Where: TDX took place in San Francisco. The platform capabilities are available globally to Salesforce customers and developers. The Agentforce Experience Layer renders across Slack, mobile, Microsoft Teams, ChatGPT, Claude, Gemini, and any MCP-compatible client.

Why: Salesforce is responding to a structural challenge: in an enterprise environment where AI agents can call APIs, invoke MCP tools, and run CLI commands directly, a platform that requires humans to navigate a graphical interface becomes a bottleneck rather than infrastructure. The initiative attempts to make Salesforce the most agent-accessible enterprise platform available, positioning accumulated customer data, workflow logic, and trust controls as a competitive advantage that AI agents working from a blank prompt cannot replicate.

Share this article
The link has been copied!