TLDR
- Salesforce’s Headless 360 decouples core business logic, APIs, and workflows from traditional browser dashboards, allowing users and AI agents to interact with Salesforce capabilities directly from external platforms such as Slack, ChatGPT or Teams.
- Permissions, security controls, and business rules automatically travel with the capability, ensuring that AI agents operating in third-party environments still adhere to established enterprise governance.
- Headless 360 marks an architectural turning point for Salesforce, says Khushwant Singh, Salesforce senior vice president of product management for its AI application development platform.
For most of its history, Salesforce customers accessed its services through a dashboard on their browsers. Now, the company is preparing for a future in which users may no longer want to open a Salesforce application at all.
That’s why it is going full force into an initiative it calls “Headless 360.” Introduced at Salesforce’s recent TDX developer conference, it lets users access its entire product ecosystem – sales, services, marketing, data, business processes and security controls – from external apps including Slack, Microsoft Teams, ChatGPT, Claude and developer tools.
It does this by essentially decoupling its core business logic, APIs, and workflows from the default browser user interface.
The strategy points to a potentially consequential shift in enterprise software: As AI agents become intermediaries between users and corporate systems, the traditional user interface may matter less.
“Headless 360 really represents an architectural turning point for Salesforce,” Khushwant Singh, Salesforce senior vice president of product management for its AI application development platform told The AI Innovator. “It’s been around for the last 27 years, and for this last 27 years, enterprise software has been a destination. … But AI is really changing that.”
With AI, employees are routinely moving among multiple applications while developers increasingly build with their preferred tools and frameworks, Singh said. Headless 360 is Salesforce’s attempt to separate more of its underlying capabilities from the Salesforce interface.
An employee working in Slack, for example, could access Salesforce information and processes without leaving the messaging app. Someone using ChatGPT could similarly access Salesforce data without going to the Salesforce application itself.
The impact of this approach is more than just convenience. If AI assistants become the primary way an employee interacts with enterprise software, vendors will have to make corporate data and actions available to agents without losing the permissions, identity controls and business rules built into the original systems.
Governance has to follow the agent
Salesforce has launched a Headless 360 Model Context Protocol server and made nearly 200 Data 360 APIs programmable, according to Singh. It has also introduced more than 100 reusable agent skills and experience layers that can display visual interfaces instead of just text.
MCP is an emerging standard for connecting AI applications with external data and tools. But exposing Salesforce capabilities to agents creates a second problem: making sure the customer’s enterprise controls still apply when users access those capabilities elsewhere.
“Our customers, they’ve spent years building their permissions, their business logic, their workflows, their identity, the governance,” Singh said. “With Headless 360, that context and governance travels with that capability as well. Customers don’t have to recreate the security models or business contexts each time they adopt a new agent or a new AI service.”
That becomes especially crucial when an AI agent acts on an employee’s behalf. Singh said customers want to see what agents are doing and restrict their permissions separately from the employees they represent.
“While that agent is working on behalf of me, I don’t want that agent to have the same capabilities that I have,” Singh said, paraphrasing a concern Salesforce hears from customers.
That distinction highlights one of the important nuances of agentic AI deployment: Giving an agent access to corporate systems does not necessarily mean it should inherit all of a user’s authority.
From scripted chatbots to grounded agents
Customer service is one area where Salesforce is seeing significant interest in agents, along with sales tasks such as lead generation, customer outreach and scheduling, Singh said.
Traditional customer-service bots generally followed predetermined decision trees: If a customer said one thing, the software followed a prescribed path. Generative AI instead allows an agent to interpret the requested outcome and determine how to accomplish it.
Singh gave the example of a customer asking to change a delivery date because no one will be home. Rather than requiring developers to anticipate a delivery date change and specifically program a path for it, an AI agent understands the goal and uses the context to determine what it should do.
For enterprises, the critical requirement is grounding the model in relevant corporate information, Singh said. That information can reside inside or outside Salesforce, with Salesforce Data Cloud bringing it together to provide context for an agent.
The distinction is important because the language generated by an AI model can be probabilistic even when the underlying customer records, permissions and business processes need to remain authoritative.
Production starts with the workflow
Singh said Salesforce sees a difference between companies successfully deploying agents into production and those conducting broad experiments without clear objectives.
More successful customers begin with defined use cases and examine existing workflows before adding AI, he said. They map processes, identify the number of steps involved, measure how long they take and determine which portions are manual. Only then do they identify tasks that could be assigned to an agent.
“There is method to the madness versus just sort of scattershot, trying everything out,” Singh said.
Salesforce is also adjusting its pricing as AI changes how customers use software. Singh said some customers still want predictable per-user, per-month pricing. Others prefer consumption-based pricing as they experiment with AI because they do not know yet how much they will use. He said Salesforce expects its offerings to remain fluid as the market develops.
“Salesforce is moving away from just being a destination you go to, but to being the trusted enterprise infrastructure that comes to you wherever you are,” Singh said. “It’s meeting the customers wherever they work and build and bringing that context, trust and governance of Salesforce today.”
Author
Get the latest insights about enterprise AI.
Subscribe to our newsletter. Thank you.
