When Salesforce announced Headless 360 at its TDX developer conference last month, the response from some quarters was underwhelmed. Pricing complaints. Questions about roadmap clarity. A few pointed suggestions that Salesforce was retreating from the CRM market.
Simon Harrison watched all of this with barely concealed frustration.
“You guys, are you kidding me?” says Harrison, founder of Actionary, a technology advisory firm. “What are you talking about, the pricing needs improving? Do you realise what’s just landed in your laps?”
Harrison’s argument is not that Salesforce has done something clever. It’s that Salesforce has done something that will set the bar for every enterprise application from this point forward, and that the UC and collaboration industry in particular needs to sit up and pay attention.
What “Headless” Actually Means
The concept is straightforward enough: strip away the user interface as the primary way of interacting with software, and expose the underlying data and functionality through an API surface that agents, command line tools, or any compatible client can call directly. Salesforce’s own framing is blunt about the direction of travel: “Why should you ever log into Salesforce again?” asked Co-Founder Parker Harris at TDX.
What makes Headless 360 significant, Harrison argues, is not the concept but the engineering behind it. Salesforce has had APIs since 2006. Redesigning those APIs specifically for non-human callers is a different undertaking entirely.
“Designing those APIs as a surface for non-human callers was a massive undertaking,” he says. “I would not be surprised if this has been going on for two-plus years.”
The result of that work is a platform with more than 60 new MCP tools and 30-plus preconfigured coding skills, enabling an agent to interact with Salesforce data precisely and repeatedly, without the ambiguity that makes most integrations unreliable in practice.
The Problem Most Vendors Haven’t Solved
Harrison built his own CRM application in Python to understand these dynamics first-hand, and the experience taught him something he says most vendors are still getting wrong.
Large language models exist in a state of constant tension between two modes: open-ended reasoning, where they are powerful and flexible, and precise action, where they need to be deterministic and consistent. Getting both from the same model, without sacrificing one for the other, is the engineering challenge that most platforms haven’t cracked.
“You go from a model where the corpus of data and the ability to pick one tool, web search, enables it to answer questions about everything it’s been trained on,” he says. “To: forget all that. Just understand intent, and use these tools.”
The mechanism that resolves this tension is Model Context Protocol (MCP), an open standard originally developed by Anthropic. It acts as a structured language between the model and the application, allowing the model to reason freely but act only through well-defined, scoped tools with clear input and output schemas.
Vendors that lack the API depth to support this are, in Harrison’s view, in serious trouble.
“There are monoliths out there that still don’t have a decent API surface. They’re going to struggle.”
What This Means for UC Platforms
For UC buyers, the most pressing question is what this shift means for the collaboration and communications tools they already rely on. The platforms sitting on the richest concentrations of workplace data, calendars, emails, conversations, and meeting recordings, are not yet doing enough with it.
Harrison’s answer is anchored in a simple daily habit he has built with his own agent-connected CRM. “Tell me what I should worry about today,” he says, describing what he asks each morning. “And my list is prioritised to maximum effect. I don’t have to traverse these various screens to figure out what I’m doing, why, and how.”
That kind of interaction should be possible within a UC platform. But it requires the platform to expose its data through an agent-ready API surface.
Microsoft Copilot in Teams is the most obvious example. MCP is now generally available in Copilot Studio, and agents in Teams channels can connect to third-party tools via MCP servers. Harrison sees the potential clearly, but also the gap between where Copilot is and where it needs to go.
“What I need from Copilot is an MCP client capability. If I could connect to my MCP server within Copilot, I’m now triaging my calendar, my emails, my chats, with the records in my CRM application, without having to look at either. And that is huge.”
Of the major UC players, Harrison reserves his strongest current praise for Zoom. At Zoomtopia 2025, the company announced MCP support within AI Companion 3.0, allowing admins to configure tools for custom and virtual agents across the platform. More recently, Zoom launched an MCP connector for Claude, pulling meeting summaries, transcripts, and scheduling data into external agent workflows. “Zoom is perhaps doing the best job in terms of the workplace agent approach,” says Harrison.
The Bar Has Been Set
Harrison is not a Salesforce partner or paid analyst. His conviction comes from a year of building agent-connected applications and observing what happens when the architecture is done properly.
His live demonstration during our conversation made the point cleanly. Within minutes, using only natural language instructions to an agent connected via MCP to his CRM, he created a fully enriched company account, pulled in a leadership team from a public web page, and received a prioritised breakdown of his most pressing tasks. No screen navigation. No manual data entry.
“I achieve things that would take days, weeks. It takes me minutes. It’s extraordinary.”
Some analysts have urged caution, noting that Salesforce has yet to disclose pricing for the headless capabilities, and that questions remain around SLAs for MCP tool calls in production environments. Those are fair points for buyers to press on. But Harrison’s argument is that the direction of travel is what matters here, not the launch-day pricing sheet.
Every application vendor, UC platform included, will eventually have to answer the same question Salesforce has spent two and a half years quietly resolving: is your API surface good enough for an agent to work with?
“The headline about Salesforce,” Harrison says, “is that this is the future of applications. It is, hands down, the only way we will decide to use applications in the future.”
Whether the UC industry moves fast enough to meet that bar is, for now, an open question.
