Salesforce launches Headless 360 to turn its entire platform into infrastructure for AI agents

Salesforce on Wednesday unveiled probably the most formidable architectural transformation in its 27-year historical past, introducing “Headless 360” — a sweeping initiative that exposes each functionality in its platform as an API, MCP device, or CLI command so AI agents can function the entire system with out ever opening a browser.

The announcement, made on the firm’s annual TDX developer conference in San Francisco, ships greater than 100 new instruments and abilities instantly accessible to builders. It marks a decisive response to the existential query hanging over enterprise software program: In a world the place AI agents can cause, plan, and execute, does an organization nonetheless want a CRM with a graphical interface?

Salesforce’s reply: No — and that is precisely the purpose.

“We made a decision two and a half years ago: Rebuild Salesforce for agents,” the corporate mentioned in its announcement. “Instead of burying capabilities behind a UI, expose them so the entire platform will be programmable and accessible from anywhere.”

The timing is something however coincidental. Salesforce finds itself navigating one of the turbulent durations in enterprise software program historical past — a sector-wide sell-off that has pushed the iShares Expanded Tech-Software Sector ETF down roughly 28% from its September peak. The concern driving the decline: that AI, notably giant language fashions from Anthropic, OpenAI, and others, might render conventional SaaS enterprise fashions out of date.

Jayesh Govindarjan, EVP of Salesforce and one of many key architects behind the Headless 360 initiative, described the announcement as rooted not in advertising and marketing principle however in hard-won classes from deploying agents with 1000’s of enterprise prospects.

“The problem that emerged is the lifecycle of building an agentic system for every one of our customers on any stack, whether it’s ours or somebody else’s,” Govindarjan advised VentureBeat in an unique interview. “The challenge that they face is very much the software development challenge. How do I build an agent? That’s only step one.”

More than 100 new instruments give coding agents full entry to the Salesforce platform for the primary time

Salesforce Headless 360 rests on three pillars that collectively characterize the corporate’s try to redefine what an enterprise platform seems like within the agentic period.

The first pillar — construct any method you need — delivers greater than 60 new MCP (Model Context Protocol) tools and 30-plus preconfigured coding abilities that give exterior coding agents like Claude Code, Cursor, Codex, and Windsurf full, dwell entry to a buyer’s entire Salesforce org, together with information, workflows, and enterprise logic. Developers not want to work inside Salesforce’s personal IDE. They can direct AI coding agents from any terminal to construct, deploy, and handle Salesforce purposes.

Agentforce Vibes 2.0, the corporate’s personal native growth surroundings, now consists of what it calls an “open agent harness” supporting each the Anthropic agent SDK and the OpenAI agents SDK. As demonstrated in the course of the keynote, builders can select between Claude Code and OpenAI agents relying on the duty, with the harness dynamically adjusting accessible capabilities primarily based on the chosen agent. The surroundings additionally provides multi-model assist, together with Claude Sonnet and GPT-5, together with full org consciousness from the beginning.

A major technical addition is native React support on the Salesforce platform. During the keynote demo, presenters constructed a completely practical companion service software utilizing React — not Salesforce’s personal Lightning framework — that linked to org metadata by way of GraphQL whereas inheriting all platform safety primitives. This opens up dramatically extra expressive front-end potentialities for builders who need full management over the visible layer.

The second pillar — deploy on any floor — facilities on the brand new Agentforce Experience Layer, which separates what an agent does from the way it seems, rendering wealthy interactive parts natively throughout Slack, cellular apps, Microsoft Teams, ChatGPT, Claude, Gemini, and any consumer supporting MCP apps. During the keynote, presenters outlined an expertise as soon as and deployed it throughout six totally different surfaces with out writing surface-specific code. The philosophical shift is important: moderately than pulling prospects into a Salesforce UI, enterprises push branded, interactive agent experiences into no matter workspace their prospects already inhabit.

The third pillar — construct agents you’ll be able to belief at scale — introduces a completely new suite of lifecycle administration instruments spanning testing, analysis, experimentation, commentary, and orchestration. Agent Script, the corporate’s new domain-specific language for defining agent conduct deterministically, is now typically accessible and open-sourced. A brand new Testing Center surfaces logic gaps and coverage violations earlier than deployment. Custom Scoring Evals let enterprises outline what “good” seems like for their particular use case. And a brand new A/B Testing API permits operating a number of agent variations in opposition to actual site visitors concurrently.

Why enterprise prospects stored breaking their very own AI agents — and the way Salesforce redesigned its tooling in response

Perhaps probably the most technically vital — and candid — portion of VentureBeat’s interview with Govindarjan addressed the elemental engineering pressure on the coronary heart of enterprise AI: agents are probabilistic techniques, however enterprises demand deterministic outcomes.

Govindarjan defined that early Agentforce prospects, after getting agents into manufacturing by “sheer hard work,” found a painful actuality. “They were afraid to make changes to these agents, because the whole system was brittle,” he mentioned. “You make one change and you don’t know whether it’s going to work 100% of the time. All the testing you did needs to be redone.”

This brittleness downside drove the creation of Agent Script, which Govindarjan described 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 capabilities as a single flat file — versionable, auditable — that defines a state machine governing how an agent behaves. Within that machine, enterprises specify which steps should observe express enterprise logic and which may cause freely utilizing LLM capabilities.

Salesforce open-sourced Agent Script this week, and Govindarjan famous that Claude Code can already generate it natively due to its clear documentation. The method stands in sharp distinction to the “vibe coding” motion gaining traction elsewhere within the trade. As the Wall Street Journal not too long ago reported, some companies are now attempting to vibe-code entire CRM replacements — a development Salesforce’s Headless 360 immediately addresses by making its personal platform probably the most agent-friendly substrate accessible.

Govindarjan described the tooling as a product of Salesforce’s personal inner follow. “We needed these tools to make our customers successful. Then our FDEs needed them. We hardened them, and then we gave them to our customers,” he advised VentureBeat. In different phrases, Salesforce productized its personal ache.

Inside the 2 competing AI agent architectures Salesforce says each enterprise will want

Govindarjan drew a revealing distinction between two basically totally different agentic architectures rising within the enterprise — one for customer-facing interactions and one he linked to what he referred to as the “Ralph Wiggum loop.”

Customer-facing agents — these deployed to work together with finish prospects for gross sales or service — demand tight deterministic management. “Before customers are willing to put these agents in front of their customers, they want to make sure that it follows a certain paradigm — a certain brand set of rules,” Govindarjan advised VentureBeat. Agent Script encodes these as a static graph — an outlined funnel of steps with LLM reasoning embedded inside every step.

The “Ralph Wiggum loop,” in contrast, represents the other finish of the spectrum: a dynamic graph that unrolls at runtime, the place the agent autonomously decides its subsequent step primarily based on what it realized within the earlier step, killing dead-end paths and spawning new ones till the duty is full. This structure, Govindarjan mentioned, manifests primarily in employee-facing eventualities — builders utilizing coding agents, salespeople operating deep analysis loops, entrepreneurs producing marketing campaign supplies — the place an skilled human critiques the output earlier than it ships.

“Ralph Wiggum loops are great for employee-facing because employees are, in essence, experts at something,” Govindarjan defined. “Developers are experts at development, salespeople are experts at sales.”

The crucial technical perception: each architectures run on the identical underlying platform and the identical graph engine. “This is a dynamic graph. This is a static graph,” he mentioned. “It’s all a graph underneath.” That unified runtime — spanning the spectrum from tightly managed buyer interactions to free-form autonomous loops — could also be Salesforce’s most essential technical wager, sparing enterprises from sustaining separate platforms for totally different agent modalities.

Salesforce hedges its bets on MCP whereas opening its ecosystem to each main AI mannequin and gear

Salesforce’s embrace of openness at TDX was hanging. The platform now integrates with OpenAI, Anthropic, Google Gemini, Meta’s LLaMA, and Mistral AI fashions. The open agent harness helps third-party agent SDKs. MCP instruments work from any coding surroundings. And the brand new AgentExchange marketplace unifies 10,000 Salesforce apps, 2,600-plus Slack apps, and 1,000-plus Agentforce agents, instruments, and MCP servers from companions together with Google, Docusign, and Notion, backed by a brand new $50 million AgentChange Builders Initiative.

Yet Govindarjan supplied a surprisingly candid evaluation of MCP itself — the protocol Anthropic created that has turn out to be a de facto customary for agent-tool communication.

“To be very honest, not at all sure” that MCP will stay the usual, he advised 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 — which now it is. A lot of people are saying that maybe CLI is just as good, if not better.”

His method: pragmatic flexibility. “We’re not wedded to one or the other. We just use the best, and often we will offer all three. We offer an API, we offer a CLI, we offer an MCP.” This hedging explains the “Headless 360” naming itself — moderately than betting on a single protocol, Salesforce exposes each functionality throughout all three entry patterns, insulating itself in opposition to protocol shifts.

Engine, the B2B journey administration firm featured prominently within the keynote demos, supplied a real-world proof level for the open ecosystem method. The firm constructed its customer support agent, Ava, in 12 days utilizing Agentforce and now handles 50% of buyer circumstances autonomously. Engine runs 5 agents throughout customer-facing and employee-facing capabilities, with Data 360 on the coronary heart of its infrastructure and Slack as its main workspace. “CSAT goes up, costs to deliver go down. Customers are happier. We’re getting them answers faster. What’s the trade off? There’s no trade off,” an Engine government mentioned in the course of the keynote.

Underpinning all of it’s a shift in how Salesforce will get paid. The firm is shifting from per-seat licensing to consumption-based pricing for Agentforce — a transition Govindarjan described as “a business model change and innovation for us.” It’s a tacit acknowledgment that when agents, not people, are doing the work, charging per consumer not is sensible.

Salesforce is not defending the previous mannequin — it is dismantling it and betting the corporate on what comes subsequent

Govindarjan framed the corporate’s evolution in architectural phrases. Salesforce has organized its platform round 4 layers: a system of context (Data 360), a system of labor (Customer 360 apps), a system of company (Agentforce), and a system of engagement (Slack and different surfaces). Headless 360 opens each layer by way of programmable endpoints.

“What you saw today, what we’re doing now, is we’re opening up every single layer, right, with MCP tools, so we can go build the agentic experiences that are needed,” Govindarjan advised VentureBeat. “I think you’re seeing a company transforming itself.”

Whether that transformation succeeds will rely upon execution throughout 1000’s of buyer deployments, the endurance of MCP and associated protocols, and the elemental query of whether or not incumbent enterprise platforms can transfer quick sufficient to stay related when AI agents can more and more construct new techniques from scratch. The software program sector’s bear market, the monetary pressures bearing down on the entire trade, and the breathtaking tempo of LLM enchancment all conspire to make this one of many highest-stakes bets in enterprise know-how.

But there’s an irony embedded in Salesforce’s predicament that Headless 360 makes express. The very AI capabilities that threaten to displace conventional software program are the identical capabilities that Salesforce now harnesses to rebuild itself. Every coding agent that might theoretically exchange a CRM is now, by Headless 360, a coding agent that builds on high of 1. The firm is just not arguing that agents will not change the sport. It’s arguing that many years of collected enterprise information, workflows, belief layers, and institutional logic give it one thing no coding agent can generate from a clean immediate.

As Benioff declared on CNBC’s Mad Money in March: “The software industry is still alive, well and growing.” Headless 360 is his firm’s most forceful try to show him proper — by tearing down the partitions of the very platform that made Salesforce well-known and alluring each agent on this planet to stroll by the entrance door.

Parker Harris, Salesforce’s co-founder, captured the wager most succinctly in a query he posed final month: “Why should you ever log into Salesforce again?”

If Headless 360 works as designed, the reply is: You should not have to. And that, Salesforce is wagering, is exactly what is going to hold you paying for it.