Salesforce debuts Headless 360 agentic platform as developers shift to probabilistic AI coding

Salesforce debuts Headless 360 agentic platform

Here comes 'enterprise vibe coding' as CRM giant aims to open development to anyone on the platform Salesforce has introduced what it calls Headless 360 at its developer event TDX, which starts today in San Francisco, designed to expand the reach of its app-building tools beyond traditional developers. The goal of Headless 360 is that everything on the Salesforce platform is now an API (application programming interface), MCP (model context protocol) server, or CLI (command line interface) command able to be called by coding agents, or by custom agents targeting specific customer requirements. The Salesforce platform includes CRM (customer relationship management), customer service, marketing, and ecommerce. The company also owns the Slack collaboration tool. Historically, the main UI (user interface) for Salesforce has been in the web browser, though it has always supported a comprehensive API. The company now prefers to talk about an "experience layer" where user interaction can be anywhere, including Slack, Teams, voice chat, ChatGPT, or a custom React application. Headless 360 means that agentic AI in any development tool - including Claude Code, Codex, Windsurf, or Visual Studio Code - can build applications that target the Salesforce platform. "The developer, the builder is talking to these tools, the tools are driving the Salesforce UI creation, configuration etc," said said Joe Inzerillo, Salesforce president of enterprise and AI technology. "We're trying to create this ecosystem where in the future most of the code is going to be written by the agents. "We see that internally in the way that we write things, already there are systems where the vast majority of code is written by the agent," he added. Agentforce Code, also known as Agentforce Vibes, is a browser-based IDE based on Visual Studio Code and is now available as part of the free Salesforce Developer Edition as well as paid subscriptions. There are new agentic integrations and it reflects the fact that some users will rely on agents to do the coding for them. Salesforce extensions, the Salesforce CLI, and organization metadata are all pre-configured. The default LLM (large language model) in Agentforce Code is Claude Sonnet 4.5. It has two modes - plan mode and act mode - and comes with a set of pre-defined agent skills for tasks such as creating custom tabs in a UI (user interface) or generating Salesforce flows. Despite the flexibility of Headless 360, Agentforce Vibes defaults to coding with Apex, the Java-like programming language which is customized for Salesforce apps. Salesforce has usage limits for its developer edition, set at 110 requests per month and 1.5 million tokens. These will refresh monthly until May 31, after which there is a final monthly allocation and no further refresh. Salesforce is keen for companies to build custom agents, rather than just using them to build applications. The definition language for this is Agent Script, which, like Apex, is customized for Salesforce. Agent Script will now be open source, Salesforce said at the TDX conference. However, the company also acknowledges that agents are "probabilistic, not deterministic," and that they "reason their way to unexpected outcomes," to quote the press release. Salesforce has a set of tools intended to control unwanted behavior, including a testing center, observability and session tracing, and the ability to define guardrails and explicit business logic for some operations via Agent Script. The company also announced the Slack Agent Kit, a collection of tools designed to bring agents on any platform into Slack with a chat UI. The story behind Headless 360 is that Salesforce has bought into the idea that most future coding will be done by agents rather than humans. The company said this is a response to customer demand. The initiative is also explicitly designed to enable non-programmers to build on the platform. "We want to expand the tent to bring in people that historically have not been part of the Salesforce ecosystem, that now are starting to build capabilities on Salesforce. Headless is a fundamental unlock that allows people to use our systems more effectively," said Madhav Thattai, EVP and GM of Salesforce AI. Most customers only use a small subset of the capabilities of the Salesforce platform, and per-user subscriptions are costly. What about the notion that with AI to assist the coding, companies can replace Salesforce with their own custom solutions built from scratch? "Maybe you could vibe yourself a SaaS application but who's going to maintain it?" said Thattai. "The notion that every company is going to build everything that they need to use themselves seems pretty crazy because it's a huge amount of expense, time, energy, and brain power that companies are focusing on things that aren't their core business." Other factors the company hopes will keep users on Salesforce include the built-in data structures and workflows that have evolved over the years for CRM, marketing, and other Salesforce use cases. Salesforce also argues that the platform's security features and trust boundaries can protect users from some of the risks of custom development, particularly when vibe coding means builders may not fully understand what they are creating. ®

Salesforce bets on conversation as the new interface for developers - SiliconANGLE

Salesforce bets on conversation as the new interface for developers Salesforce Inc. wants to make conversation the interface for developers, bringing agents to the foreground with a "headless" version of its platform designed to let humans and artificial intelligence build from nearly anywhere. The new capacity, dubbed Headless 360, will usher in a new paradigm for the company. Every capability in Salesforce's platform is accessible via application programming interfaces, Model Context Protocol tools for connecting models to data and tools, or the command line. That means humans and artificial intelligence models alike can build on them and compose apps across platforms with far fewer constraints. According to the company, this opens a whole new horizon. It also means developers can bring their favorite coding agents from any vendor to the party, including Claude Code, Cursor, Codex and Windsurf. AI also opens up the possibility of prompting ideas and generating code from within chat interfaces such as Slack, mobile, ChatGPT, Claude, Gemini, Teams or any client that supports MCP. To support those possibilities, Salesforce introduced a new way to build agents with Agentforce Vibe 2.0, an artificial intelligence builder that brings agent development natively into the Salesforce platform with multimodel support, including Claude Sonnet and GPT-5. It also provides a development partner that understands business logic, not just code. On top of that, developers will gain full control over the user interface, including native React support, a popular open-source JavaScript library designed for building user interfaces. That will allow them to create fully custom experiences from scratch that are fast, interactive and lively. The company said one enterprise customer saw Agentforce adoption jump from 22% to 78% in six weeks. The agent did not change; the experience did. In addition to evolving Vibe, Salesforce is also launching Testing Center, which is designed to reveal logic gaps, policy violations and inconsistent outputs before they reach users. With Custom Scoring Evals, developers can measure how something performed, not simply whether it ran or failed. They can set a series of standards and evaluate how well those standards were met, rather than relying only on pass-fail criteria. After launch, Observability and Session Tracing will make it possible to see not only what happened in production, but why. When an agent drifts off-task, Salesforce said it will help teams get to the root cause in hours rather than weeks. A/B testing is also rolling out, allowing multiple agent versions to run against real traffic at the same time for more data-driven testing and decision-making. In addition to the company's agentic AI announcements, Salesforce also announced a major expansion to Agent Fabric, the company's trusted agent control plane for companies building multivendor agentic networks. Agent Fabric now brings automated discovery, agent authoring and centralized large language model governance for orchestration. That allows handoffs, model selection and decision-making to happen within enterprise boundaries, with cost and risk optimized without sacrificing speed or quality. Salesforce said the new paradigm for controlling multivendor agent workforces often comes down to having a single "lead" agent that orchestrates the broader labor force. In practice, that can mean a central model overseeing handoffs, coordinating smaller models for reliability and speed, managing costs and token limits, and identifying optimized routes.

Salesforce TDX 2026 - headless thinking for the Agentic Enterprise era as developers face the shift from determinism to probabilism

When Salesforce co-founder Parker Harris asked earlier this month why anyone should ever log into Salesforce again, it was a comment that caught the eye. It was also unlikely to be the end of the story, of course, as attendees at this week's TDX developer conference in San Francisco are about to find out. Meet Headless 360, the follow-through on Harris' tease. This is based on the first premise that for the past quarter of a century, using Salesforce meant working inside Salesforce. But with today's push towards the Holy Grail of the Agentic Enterprise, that's a limiting factor, reckons the firm, going so far as to assert that if a platform requires humans to click through UIs or write code directly to make progress, it is not ready for the Agentic Enterprise. As per the launch blah blah, Salesforce Headless 360 provides: ...the capabilities your agents need most, exposed as an API, MCP tool, or CLI command so humans and agents can build, act, and deliver experiences on any surface. Instead of burying capabilities behind a UI, expose them so the entire platform will be programmable and accessible from anywhere It also brings three new innovations: ● New MCP tools and coding skills that give your coding agent full access to your platform. ● A new experience layer that renders rich, native interactions across every surface, from Slack to Voice to WhatsApp. ● New tools that give you control over how agents behave in production, before launch and after. The life of a developer has come a long way since he learned waterfall development, says Joe Inzerillo, President of Enterprise and AI Technology at Salesforce. That meant big, long planning cycles and detailed specifications before you ever broke ground. It was slow and slow doesn't fit today: Now we've got to the era of agentics, the speed limit is insane. We're now going really fast. What used to be weeks is now in some cases hours. We have digital labor and humans doing things. There's also a big shift going on that will get a lot of air time over the next two days at TDX, that between determinism and probabilism, he argues: Typically the vast majority of developers, the vast majority of trailblazers, were developing deterministic experiences. You can put in some input, you can define how you want that input modulated or enhanced, and it got some output, You could do that every day, and one plus one was always equal to two, no matter what you did, no matter what day you did it on. It was incredibly deterministic. But we start to talk about agents, and agents are not software. Agents are something else. They're also not human. We have to be careful not to completely anthropomorphize them, but what they are is a much more probabilistic, stochastic system that means that you can't necessarily put in the exact same input and get the exact same output every time. That requires different tooling, a different development cycle. It's really important to bear that in mind as we start to think about building deterministic experiences, building probabilistic experiences, and we're going to be using AI and agents to do both of those things. Hence, the Headless 360 platform, he says: This is really leaning into the fact that coding engines have just gotten really, really good. You can sit down and give [one] a detailed list of instructions and have it go to town writing code for you. It used to be that a lot of processes were [restricting] development, or administrative resources. Now the limit is almost your imagination. It's humans and agents working together to build those deterministic systems like we have in the past, and human and agents working together to build other agents, which is another experience in the mix. But it's really important to have the core values and capabilities of Salesforce integrated into all this, he adds: It's not just tools it's actually more of a development environment and a development philosophy that we're also trying to export to our partners and Trailblazers. If we look at the way that you would typically develop code in a deterministic flow, you generally build, you then evaluate, make sure it's good, you deploy it, and then you make sure it's good in production. That's kind of the cycle. There's a bunch of tooling that we have now released that's either in pilot or going GA right now to help support these functions, building with agents, having agents help you build an agentic experience, and not just our agents, but the whole universe of agents. But when you turn your attention towards building a agentic experience, that. set of tools isn't sufficient. You need other tools that are really designed for the development of agentic experiences. He cites experimentation and control-and-orchestrate as two big areas that are new ground to be covered: When you think about probabilistic experiences, it's really hard to know just in the testing phase whether or not this agent is actually getting better or worse when you're making changes. So part of the way that you have to do that is you test your way into it with A/B experimentation. So, here's my agent in production, here's my agent that I made a tweak on, is that agent better or worse? You really need to see that happen in production, with real traffic and real humans that that agent is talking to, to know that it's good. When you're doing the observe cycle, you also have to control and orchestrate the agents. Ultimately, it's not going to be a single agent that's in your organization or serving your enterprise. It's going to be multiple agents. Understanding how those agents interact is a super-important part of the agentic development life cycle. We can't just ship these agents and hope they work together. We have to think about how they're going to talk together. When this agent gets a question that it doesn't know the answer to, how does it talk to another agent? When people are trying to do a job, what agent should they be going to? How do we want to put that front and center in front of folks? He concludes: We need to let people build the way they want to build because everybody's going to have a slightly different set of requirements. Everybody's going to have a slightly different set of working, so we want to put out tools that help, that guide people into a life cycle that we know will be successful. But we also want to inter-operate. Ian Thomas is on the ground at TDX this week and he'll be reporting back on the vibe from the developer community to this shift in messaging from Salesforce. More to come.