TL;DR
On April 15, 2026, at TDX 2026 in San Francisco, Marc Benioff and the Salesforce team announced what the company itself called
"the most ambitious architectural transformation in our 27-year history."
The name: Headless 360.
The shape of it, in three lines:
- Every Salesforce platform capability is now exposed as an API, MCP tool, or CLI command
- AI agents can work with Salesforce without ever opening a browser
- The customer experience can show up anywhere — Slack, Microsoft Teams, ChatGPT, Claude, Gemini, mobile
Day one ships with 100+ tools and skills, including 60+ MCP servers, 30+ preconfigured coding skills, and Agent Script (a new DSL that's now open-sourced).
For non-Salesforce shops, this still matters. It's the clearest signal yet that enterprise SaaS is changing shape, and the teams that adapt early will win the next contract.
The day Salesforce killed its own browser
Twenty-seven years ago, Salesforce was born from one simple idea: software shouldn't live on your machine. It should live in the browser.
"No Software" became a slogan, then a culture, then a $250B market cap, then the world's #1 CRM.
Then on April 15, 2026, Salesforce stood up at TDX and basically announced:
"The Salesforce UI is now optional."
It sounds contradictory at first. Why would the company that built a brand on browser-first software walk away from it?
The answer isn't that Salesforce is being reckless or pivoting at random.
The answer is one uncomfortable truth: enterprise users don't live inside the Salesforce UI anymore.
They live in Slack. In Microsoft Teams. In ChatGPT. In Claude.
And more importantly — the AI agents that are starting to do their work don't have a browser to begin with.
This is the moment Salesforce decided to stop fighting the current and start steering it.
Headless 360, in three lines
Simplest possible framing:
- CRM, marketing cloud, ecommerce, service cloud — every Salesforce surface is now an API + MCP tool + CLI command
- AI agents like Claude Code, Cursor, Codex, and Windsurf can hit Salesforce data and workflows directly, no browser involved
- Customer experience can render on Slack, mobile, Teams, ChatGPT, Claude, Gemini, or any MCP-compatible client
What's beautiful here isn't that any of this is technically possible. Salesforce APIs have existed for years.
What's new is that it's complete, opinionated, day-one shipped, and aimed squarely at the AI agent generation that's about to dominate.
That's the difference between "having an API" and "being an API platform."
Why now — beyond the AI hype
On the surface this looks like Salesforce chasing the agent moment. But there are at least four structural reasons it had to happen.
1. Users aren't in Salesforce anymore
Sales reps live in Slack all day. Customer success lives in Zendesk. Marketing lives in Notion.
"Going into Salesforce" became a context switch everyone tried to avoid.
If Salesforce doesn't show up where the user already is, the user starts looking for a CRM that does.
2. AI agents need structured access, not screen scraping
If you point Claude Agent at Salesforce through a browser, it's:
- Slow
- Brittle every time the UI changes
- Not scalable
- Likely to misread context
Through an MCP tool, the agent understands the schema, the relationships, the permissions — in one prompt.
3. Browser UI is a bottleneck in the agentic era
If 30% of work in the near future gets done by agents (and likely 50% within five years), a UI designed for humans clicking is just overhead.
4. Competitors are closing in
Microsoft Copilot Studio + Dynamics 365 has real momentum. ServiceNow's Now Assist is eating into enterprise workflow share.
If Salesforce hadn't moved first, it would have started looking like legacy in the eyes of AI-native CIOs.
Three things developers actually need to know
1. 60+ MCP tools — Claude Code, Cursor, Codex, Windsurf get direct access
This is the part developers got most excited about.
Open Claude Code, install the Salesforce MCP servers, and type:
"Look at my Salesforce pipeline and put together a deck comparing Q1 vs Q2."
The agent will:
- Query the Opportunity object
- Aggregate by stage
- Compare periods
- Generate the slides
All in one conversation. No opening Salesforce. No CSV export. No hand-written SOQL.
Critically, MCP support spans Claude Code, Cursor, Codex, Windsurf, and any MCP-compatible client — so developers don't get locked into one vendor.
2. Agentforce Experience Layer — UI that renders across surfaces
This is the smartest piece, in my read.
The core idea: separate "what an agent does" from "how it appears."
Developers define the agent behavior once, then Salesforce renders rich interactive components across:
- Slack
- Mobile apps
- Microsoft Teams
- ChatGPT
- Claude
- Gemini
- Any MCP-compatible client
That cuts development time from months (building six different UIs) to a single definition.
The deeper philosophy: push the agent experience into the customer's workspace, instead of dragging the customer into the Salesforce UI.
That's a paradigm flip, not a feature.
3. Agent Script — open-source DSL for deterministic agents
The piece most people overlook, but it matters.
Agent Script is a new domain-specific language for defining agent behavior in a deterministic way.
Why deterministic? Because in the enterprise, an agent that "does it this way today, maybe a different way tomorrow" doesn't pass compliance, doesn't pass audit, doesn't pass risk review.
Agent Script lets developers:
- Define predictable flows
- Test them
- Audit them
- Reproduce them
And the kicker — it's open-sourced, generally available. Anyone can implement their own runtime.
That's a clear signal Salesforce is trying to make Agent Script an industry standard rather than another proprietary lock-in.
