Salesforce Says the Future of Enterprise AI is ‘Headless’

The future of enterprise software may not have a front-end. That is the vision Salesforce is betting on as organisations increasingly deploy AI agents across their operations.

Instead of employees constantly switching between dashboards, applications, and analytics tools, Salesforce is pushing a ‘headless’ enterprise model where insights, recommendations, and actions are delivered directly within the workflow.

Earlier this year, the company launched Salesforce Headless 360, an API-first architecture that allows developers and AI agents to access Salesforce capabilities through APIs, MCP tools, and CLI commands without using the traditional user interface.

“The advantage of ‘headless’ is that the context is being built by platforms such as Informatica and MuleSoft together, while the insights are delivered in the flow of work where the business user is,” Romil Agarwal, VP, Sales, Salesforce India, tells AIM in an exclusive interview.

Informatica, a data integration and management platform, and MuleSoft, Salesforce's application integration platform, help enterprises connect, unify, and move data across disparate systems. Citing an example, Agarwal explains that business users can access CRM information directly within applications such as Slack, eliminating the need to switch between multiple interfaces.

Moreover, integration technologies, APIs, and protocols such as MCP are poised to turn critical as enterprises move deeper into the era of agentic AI.

“The API-led economy is only going to get much stronger now when organisations are looking at integrating their enterprises,” Agarwal notes.

While AI agents may sit at the front-end of business processes, APIs remain the mechanism through which enterprise applications exchange information. Without those connections, agents lack the context required to perform meaningful work.

MuleSoft Anypoint Platform

Salesforce is bringing the MuleSoft Anypoint Platform to Hyperforce in India, allowing local deployment of its integration platform. 

MuleSoft Anypoint Platform is a unified, enterprise-grade integration and API management platform. It allows businesses to securely design, build, deploy, and manage APIs and integrations, connecting SaaS, on-premises, and legacy systems.

The move will help enterprises meet data residency requirements, which are closely tied to the needs of industries, including financial services, healthcare, telecom, and the public sector, where regulatory obligations around data handling continue to increase.

The importance of local deployment has only grown following India’s Digital Personal Data Protection (DPDP) Act. Agarwal says that local data residency allows organisations to meet compliance requirements while ensuring that sensitive information remains within approved jurisdictions.

He points to use cases such as home loan processing, where organisations handle highly sensitive personal information, including Aadhaar, PAN, and property documents. Keeping such workloads within India becomes increasingly important as enterprises automate document-intensive processes using AI.

The Missing Ingredient

Agarwal believes the industry’s struggle to move AI projects from pilot stages into production is often less about the quality of AI models and more about the quality of enterprise data infrastructure.

Referencing industry reports that suggest many AI pilots fail to scale, he argues that the challenge is rarely the AI itself. “The challenge is the context,” he says.

Creating that context requires organisations to address multiple layers of data management, including extraction, classification, cataloguing, and lineage tracking. Without these foundations, AI agents may have access to information but lack the ability to understand how different datasets relate to one another.

For enterprises operating a mix of legacy systems, modern cloud applications, and multiple AI agents, orchestration becomes increasingly important. 

Agarwal says Salesforce’s MuleSoft Agent Fabric is built to tackle this challenge and helps organisations coordinate information flows across different agent platforms and applications.

He highlights financial services as one sector where integration and automation are already delivering measurable results. Meanwhile, at Hero FinCorp, the company claims to have reduced loan processing times to 30 minutes by combining Agentforce on the front end with MuleSoft-powered integrations on the back end.

Loan approvals often require interactions across multiple systems, including banking networks, government databases, KYC verification platforms, and account validation services. By connecting these processes through APIs and automation, organisations can dramatically reduce manual effort and turnaround times.

The same principles can be applied to intelligent document processing, where AI can automatically read, classify, and extract information from both structured and unstructured documents. According to Agarwal, this is where the real value of enterprise AI lies. AI models may generate responses, but they still depend on connected systems and trusted data to produce useful outcomes.

As companies deploy more AI agents across their operations, the focus is likely to shift from the models themselves to the infrastructure that powers them. For Salesforce, that infrastructure is built on APIs, integration platforms and data layers that provide the context AI needs to act.