On April 15, 2026, at TrailblazerDX in San Francisco, Salesforce did something it has been threatening to do for years. It took the browser out of the equation entirely. The question is whether that matters for your org, your team, and your roadmap.
Before we get into what Salesforce Headless 360 actually is, let me clear up the most common source of confusion that has been circulating in the community since the announcement. If you have worked with Commerce Cloud in the last few years, the word “headless” means something specific to you: a composable storefront where you own the front end and Commerce Cloud handles the back-end APIs. That still exists. That is not what this is.
Salesforce Headless 360 is something else entirely. It is the company’s answer to a much larger question: what does an enterprise platform look like when AI agents, not people, are its primary users?
1. What Salesforce Headless 360 Actually Is
The official one-liner from Salesforce is blunt: “Everything on Salesforce is now an API, MCP tool, or CLI command, and agents can use all of it.” Marc Benioff put it even more directly on the day of the announcement: “No Browser Required. Our API is the UI.”
For 27 years, Salesforce has been built around humans logging into a browser, navigating point-and-click interfaces, and clicking Save. Every integration that came before this required custom scaffolding: authentication setup, schema translation, rate-limit handling, error wrangling. You did this once per integration, and then again when the next one came along.
What Headless 360 does is expose the entire stack — Salesforce CRM, Agentforce, Data 360, and Slack — as a single programmable surface that any AI agent, coding tool, or external system can reach without a browser session. Not as a workaround. As the intended first-class path.
Why Now
Jayesh Govindarjan, EVP at Salesforce and the lead architect behind Headless 360, told VentureBeat that the re-architecture began roughly two and a half years ago when Salesforce leadership made a deliberate bet to rebuild the platform for agents. What they are shipping now is what their own internal field delivery engineers needed first — battle-hardened tooling they built for themselves, then productized for customers.
Salesforce co-founder Parker Harris framed it with a rhetorical question from the TDX stage that is worth sitting with for a moment: “Why should you ever log into Salesforce again?” That is either a compelling vision or a slightly alarming statement, depending on how much of your job currently involves logging into Salesforce.
2. The Three Pillars
Salesforce structured the announcement around three core deliverables. Over 100 tools and capabilities shipped on day one, but these three pillars are the architecture.
The numbers on day one are worth noting on their own.
3. Under the Hood: MCP, APIs, CLI, and the Experience Layer
If you are an architect or developer, the technical specifics matter. Here is how the three programmable surfaces actually work.
The MCP Server
The Salesforce DX MCP server — @salesforce/mcp — is the most genuinely net-new piece of the announcement. It exposes named toolsets that a coding agent can call without writing custom integration code. Those toolsets include:
| Toolset | What It Gives Agents Access To |
|---|---|
| metadata | Deploy and retrieve metadata between DX projects and orgs |
| data | Execute SOQL queries against live org data |
| testing | Run Apex tests and agent evaluation tests |
| lwc | Create LWC components, Jest tests, accessibility guidance |
| code-analysis | Static analysis via Salesforce Code Analyzer |
| devops | CI/CD pipeline access via DevOps Center MCP |
| aura | Migration blueprints and LWC transition guidance |
You install it with a single command, point it at one or more orgs, and specify which toolsets to load. Once it is running, any MCP-compatible client — Claude Code, Cursor, OpenAI Codex, Windsurf, VS Code — picks up the tools automatically.
npx -y @salesforce/mcp \
–orgs myDevOrg \
–toolsets metadata,data,testing,lwc
Salesforce MVP Amit Chaudhary from Apex Hours put it well when he said this is “partly a reframing of APIs that already existed, but the MCP layer is genuinely net-new.” The APIs have been there since 2006. Having them organized into named toolsets that a coding agent can call without any custom integration scaffolding — that is new.
The CLI: Agent Commands
New sf agent commands make the lifecycle fully scriptable. You can generate an agent spec from plain language, deploy it, run evaluations, and trace sessions — all from a terminal, all without touching a UI. JWT OAuth flow enables truly headless authentication for CI/CD pipelines.
Agent Script (Now Open Source)
One of the more interesting moves at TDX 2026 was Salesforce open-sourcing Agent Script, its structured language for defining agents. The full grammar, parser, and compiler landed on GitHub at github.com/salesforce/agentscript. Whether this becomes a broader standard or stays a Salesforce-specific DSL depends on adoption, but open-sourcing it sends a signal about the direction.
The Agentforce Experience Layer (AXL)
AXL is worth a separate explanation because it solves a problem that has been invisible until now. If you have built agents, you have probably noticed that the logic you write for one channel cannot easily be reused on another. The agent that works in Slack needs to be rebuilt for Teams. The experience that renders well on mobile looks wrong on desktop. AXL separates what an agent does from how it renders.
You define rich interactive components — booking flows, approval cards, data layouts, decision tiles — once. AXL renders them natively on Slack (Block Kit), Microsoft Teams, mobile, ChatGPT, and any other MCP-compatible surface. Salesforce describes it as a “control plane for how AI shows up across your business.” It is included in the Agentforce 360 Platform license.
Trust Layer: Still There, Still Non-Negotiable
The Einstein Trust Layer applies to every Agentforce session whether it comes through a browser, an MCP call, or a CLI command. PII masking, dynamic grounding against org data with field-level security checks, and zero-data-retention agreements with LLM providers are all enforced. You cannot configure this off. Agent API calls also count against your daily org API quota — Enterprise Edition orgs start at 100,000 REST calls per 24-hour rolling window, which can deplete fast under agent workloads.
4. Where Headless 360 Sits in the Broader Platform
Salesforce positioned Headless 360 as the programmable access layer sitting over four existing platform layers. It is worth understanding the stack before you start planning integrations against it.
| Layer | Product | Role |
|---|---|---|
| System of Context | Data 360 | Trusted, unified, real-time business data |
| System of Work | Customer 360 | Decades of business logic orchestrated by agents |
| System of Agency | Agentforce | Build, deploy, and govern agents at scale |
| System of Engagement | Slack | Where humans and agents work together |
Slack’s positioning as the “front door to the Agentic Enterprise” is not incidental here. The 300% growth in custom agent deployments on Slack since January 2026 gives Salesforce a real data point to lean on. The vision is that you might never open salesforce.com, but you will interact with Salesforce constantly through your Slack workspace.
5. What This Means for You, Depending on Your Role
The honest answer is that Headless 360 means very different things depending on who you are and what you do day-to-day.
- MCP server is installable today, no waitlist
- Claude Code, Cursor, Codex, Windsurf all compatible
- Agentforce Vibes 2.0 IDE with your choice of model
- Agent Script is open source and learnable now
- Watch your daily API limits under agent workloads
- Design for tightly scoped OAuth — avoid full access
- Plan API governor limits as a real constraint
- AXL changes your channel strategy fundamentally
- Session Tracing and Scoring Evals are production tools now
- MCP SLAs are not yet documented — push Salesforce on this
- The keynote was honest: this was a pro-code announcement
- Day-to-day clicking reduces; reviewing Session Traces increases
- Flow Builder and declarative tooling still exist — this is additive
- Governance becomes more important, not less
- Upskilling on agent evaluation concepts is worth your time now
The “builder gap” concern raised by several community voices after TDX is real and worth acknowledging. The declarative ecosystem — the admins and business analysts who built the majority of Salesforce customizations over the past decade — was not the primary audience for this keynote. That tension is not new to Salesforce announcements, but it is more visible here than usual.
6. Who Is Already Doing This
Salesforce featured real customer examples at TDX. Two are worth looking at closely because they give you a realistic picture of what is achievable today versus what is still aspirational.
Engine built its customer service agent, named Ava, in 12 days using Agentforce. It now handles 50% of customer cases autonomously across five agents covering customer-facing and employee-facing functions. Data 360 provides the underlying infrastructure; Slack is the primary workspace. CEO Elia Wallen described it as deploying production-ready AI in less than two weeks while driving measurable savings in operational spend.
Indeed is using coding agents with live access to its Agentforce platform directly within their existing development tools. Oliver Bodden, Senior Product Manager, described the primary benefit as reducing the gap between idea and implementation — going from concept to working code faster because the agent has full context about the org without needing it to be re-supplied each time.
7. The Questions Salesforce Has Not Answered Yet
This is the part of the announcement that deserves more scrutiny before you make architectural decisions around it.
On Pricing
Headless 360 does not have its own SKU or separately disclosed price. It is described as capabilities exposed across existing licensed products. Scott Bickley from Info-Tech Research Group put the issue directly: Salesforce’s typical approach is to announce capabilities and then land the SKU later. CIOs and architects should ask about pricing explicitly before building architectural dependencies on features that might arrive in a premium cost tier.
On Per-Seat Pricing in an Agent-First World
This is the deeper structural question. Enterprise software has been priced per seat for decades because a seat meant a person logging in. Agents do not log in. They make API calls. What does a per-seat CRM look like when the primary users are not humans? Salesforce did not answer this at TDX, and several community commentators flagged it as the most important unresolved question from the entire announcement.
On MCP as a Standard
Govindarjan was candid with VentureBeat when he said he is “not at all sure” whether MCP will remain the standard. Salesforce is offering API, CLI, and MCP as parallel surfaces and is not betting exclusively on MCP. For long-term architecture decisions, that hedging is something to keep in mind.
8. Agentforce Vibes 2.0 and What Is Available Now
Separate from Headless 360 but announced at the same TDX keynote, Agentforce Vibes 2.0 is the development environment story that developers will interact with most directly. It is a multi-model, org-aware browser-based IDE built on VS Code. The default model is Claude Sonnet on the Pro tier, with GPT-5 mini as the Core fallback when monthly limits are reached. You can also connect your own model preference.
Salesforce claims the build loop is up to 40% faster because of reduced context-switching between tools. Developer Edition orgs include access through May 31, 2026, at which point paid billing begins. The allocation is 110 requests per month with 1.5 million tokens, refreshing monthly — but only through that May 31 cutoff. After that, the system moves to Flex Credits, flat-fee, or per-user licensing depending on your usage pattern.
Also worth bookmarking: labs.agentforce.com now gives you instant Agentforce access from any IDE with no org provisioning required. For early exploration, that is a meaningfully lower barrier than it was before TDX.
9. How This Compares to Previous “No More Browser” Moments
Salesforce has announced this kind of architectural shift before. Mobile First came in 2012 with Salesforce1. Slack First came after the 2021 acquisition. In both cases, the browser never actually went away — it gained a new companion surface. So the natural question is whether Headless 360 is different.
There are a few reasons to think this time has more structural weight. The MCP layer is genuinely new tooling, not a rebranding of existing integrations. AXL as a primitive for multi-surface rendering does not have a predecessor in the Salesforce stack. And the agent lifecycle controls — Testing Center, Scoring Evals, Session Tracing — acknowledge something that previous announcements glossed over: agents are probabilistic, not deterministic, and the platform needs to be built for that.
Neil Raden at Diginomica read the announcement at an operating model level, which I think is the right frame: Salesforce is not just shipping features. It is re-optimizing the platform for agents as primary users rather than people, and betting that enterprise value flows to whoever owns the substrate those agents run on.
Headless 360 is the most significant architectural signal Salesforce has sent in years.
The MCP tooling is real, the agent lifecycle controls address genuine production concerns,
and the AXL multi-surface rendering solves a fragmentation problem that every team building
agents has already felt.
But before you restructure your development workflow around it, get clear answers on three
things: what the pricing model looks like once the billing transition completes, what SLAs
Salesforce will commit to for MCP tool calls, and how agent consumption fits into your
existing API governor limit posture.
The strategic direction is right. The tactical details still need to be negotiated.
Start experimenting with the MCP server in a sandbox today. Hold off on production
architectural commitments until the commercial model is clearer.
10. Further Reading
- Salesforce: Introducing Salesforce Headless 360 — No Browser Required
- Apex Hours: Salesforce Headless 360 — The Entire Platform Is Now an API
- VentureBeat: Salesforce Launches Headless 360 to Turn Its Entire Platform Into AI Infrastructure
- SalesforceDevops.net: TDX 2026 — Salesforce Goes Headless and Widens the Builder Gap
- The Register: Salesforce Debuts Headless 360 Agentic Platform
Disclaimer: This blog post contains AI-generated content. While the information has been reviewed for accuracy, please verify key details independently before making architectural or business decisions based on this material.
Discover more from DYDC
Subscribe to get the latest posts sent to your email.