SAN FRANCISCO, April 20, 2026, 08:55 PDT
- Salesforce has launched Headless 360, exposing platform functions as APIs, MCP tools and command-line actions for AI agents, with more than 60 new MCP tools and 30-plus coding skills available now.
- The rollout lands as software companies try to answer investor fears that AI could weaken traditional software pricing and business models.
- Salesforce says Agentforce ARR reached $800 million in fiscal 2026, while Fortune reported the company used AI agents internally to cut annualized support costs by $100 million.
Salesforce has launched Headless 360, a bid to let AI agents work across its software without a human opening the CRM in a browser. Unveiled at the company’s TDX developer conference in San Francisco, the package turns Salesforce functions into APIs, Model Context Protocol tools and command-line actions that software agents can call directly.
The timing matters. Software stocks were hammered earlier this year as investors questioned whether fast-moving AI tools could undercut the old seat-based software model, and Salesforce is still under pressure to show Agentforce can move from pilots into everyday production. Reuters reported in February that the software and services index had lost about $1 trillion in market value during the selloff, while analysts said Salesforce needed to prove broader enterprise adoption.
The field is getting crowded, too. OpenAI launched Frontier in February to help companies build and manage agents on top of existing infrastructure, and Adobe on Monday rolled out CX Enterprise, its own AI-agent suite for marketing and customer work, a sign that big software groups are remaking their stacks around automation at speed.
Salesforce said Headless 360 has three main pieces: new developer tools, a new experience layer that can render work across surfaces such as Slack, Voice and WhatsApp, and governance features to test and tune agents before and after launch. MCP, first introduced by Anthropic, is an open standard for connecting AI applications to outside tools and data — basically a common way for an agent to call software instead of asking a person to click through it.
More than 60 new MCP tools and 30-plus preconfigured coding skills are available immediately, Salesforce said, giving coding agents in tools such as Claude Code, Cursor, Codex and Windsurf live access to customer data, workflows and business logic. VentureBeat said the broader package includes more than 100 tools and skills, while CIO quoted analysts saying Salesforce is trying to become the layer that manages agent-led work even outside its own interface.
Salesforce and its customers are leaning hard on proof points. Engine, a business travel platform, said it built its Eva customer-service agent in 12 days using Agentforce and now uses the software to handle 50% of chat cases with no human intervention, cutting average support handling time by 15% and sales research time by 40%, the company said in a press release.
Fortune reported over the weekend that Salesforce has used Agentforce internally to handle 3 million customer conversations and cut annualized support costs by $100 million. Jim Roth, Salesforce’s president of customer success, told the magazine the “real unlock” was capacity as much as cost, a line that goes to the heart of Salesforce’s new pitch to investors: AI should expand revenue opportunities, not just trim expense. Fortune
The company has some numbers behind that argument. Salesforce said fiscal 2026 revenue rose 10% to $41.5 billion, Agentforce ARR climbed 169% year over year to $800 million, and cumulative Agentforce deals reached 29,000. It forecast fiscal 2027 revenue of $45.8 billion to $46.2 billion and said it expects organic growth to reaccelerate in the second half.
Marc Benioff has framed that push as a rebuttal to “SaaS-pocalypse” fears. Reuters reported in March that software executives were fighting back against the idea that AI models could wipe out traditional business applications, and said Salesforce’s defense rests in part on its store of proprietary customer and workflow data, which is harder for rivals to replicate from scratch. Reuters
But there is still a real gap between the pitch and the proof. Rebecca Wettemann, chief executive of Valoir, told Reuters in February Salesforce still needs to show customers are moving agents from pilots to production at scale, while CIO quoted Info-Tech analyst Scott Bickley warning buyers to press the company on pricing before they build dependencies around Headless 360. That is the risk now: Salesforce may widen its reach, but only if enterprises accept the cost, governance and vendor-lock-in trade-offs that come with turning more of their work over to autonomous software.
