AI & Facilities Technology
Insights from a conversation between Vixxo IT leadership and PolyAI on giving AI agents "hands" inside a facilities management business.
Most companies chasing AI adoption run into the same wall: the models are genuinely impressive, but they can't actually do anything inside the systems a business runs on. As Vixxo VP of IT Bobby Hunnicut put it in a recent conversation with PolyAI, today's AI has incredible brains and no hands. Model Context Protocol, or MCP, is how you give it hands — and it's already reshaping how Vixxo operates.
Vixxo CTO Derek Neighbors and Hunnicut sat down with Damian Sasso, Group Product Manager at PolyAI, to talk through how Vixxo is rolling AI agents out to employees across the company, not just engineering.
Key Takeaways
| • | MCP or CLI access is now a pass/fail evaluation criteria for any new Vixxo vendor. |
| • | Agent access is scoped to the individual employee's existing credentials — no separate, wide-open service accounts. |
| • | Simple data-shuffling gets automated cheaply; AI spend goes toward decisions that actually require reasoning. |
| • | An agent with 90 days of asset history caught a plumbing issue disguised as a broken beverage machine — built by an operations employee, not an engineer. |
Facilities management isn't the industry people picture when they think about cutting-edge AI. It's a "bluecollar, turn the wrench" business, as Neighbors describes it — HVAC units, refrigeration, coffee machines, and a lot of hardware that predates modern software entirely. Vixxo services that equipment across thousands of retail, convenience store, and QSR locations, and none of the systems surrounding that work were built to be agentic.
That's the gap Hunnicut is pointing to. A powerful AI model with no way to touch a work order system, a dispatch tool, or an invoice queue can talk about the problem, but it can't solve it. Giving agents real, governed access to the systems employees already use every day is what turns a chatbot into something that changes how work actually gets done.
Hunnicut described a shift in how Vixxo evaluates any technology vendor: does it offer an MCP or a CLI? If not, that's a serious strike against it, no matter how strong the underlying product is. Software without one of those integration points forces employees to leave the AI harness they're working in and go click around in a separate dashboard — friction that undermines the entire point of giving people an agent in the first place.
"Try and do your entire job without having to leave the harness. If you have a solution that does not work with that, that's probably an inflection point for us to discuss."
— Bobby Hunnicut, VP of IT, Vixxo
Internally, Vixxo has built what amounts to a skills hub — a shared library where anyone who builds a skill for their AI harness can publish it, and anyone else in the company can pull it down, fork it, or adapt it. An operations employee's dispatch logic can become the starting point for someone in a completely different part of the business.
Neighbors was clear that MCPs and CLIs don't replace REST APIs — they build on top of them. A well-structured REST API, with clean, predictable resources, is what makes it fast for engineers to build a good MCP or CLI on top of it. A messy underlying API produces a messy MCP; a clean one lets an agent explore intelligently and often figure out how to use it with little documentation at all. The API remains the foundation. What's changing is how few people ever need to touch it directly.
One of the sharper points in the conversation was a warning against the most common way companies waste AI spend: using an expensive model to do simple, repetitive data-shuffling that doesn't require any judgment at all. Neighbors used invoice processing as the example — a task that used to involve someone manually moving files between folders, copying data into a spreadsheet, and feeding it into another system. Automating that is worthwhile, but it's automation, not intelligence, and Vixxo's engineering team treats it accordingly: let AI prove the workflow works, then rebuild the proven steps as lightweight, cheap automation and free up the token spend for problems that actually need reasoning.
The Dispatch Story
A Circle K's ice dispenser goes down. Default routing sends a beverage technician — same as always.
An agent with 90 days of asset history at that location flags something the default routing can't see: two clogged toilets at the same site in the last 60 days, and a second beverage-machine call that looks related.
Real diagnosis: a plumbing clog downstream. A plumber goes out instead — with a camera, a water jet, and a second person on the truck, so there's no callback. What would have been two or three separate service calls becomes one. The logic behind it was built by an operations employee, not an engineer.
Governance came up directly, and Neighbors laid out how Vixxo approaches it: rather than issuing a single wide-open key or token to an MCP, access is scoped to the individual user's existing credentials, typically through OAuth. An employee's agent inherits exactly the permissions that employee already has — nothing more. If someone can only see certain data through the UI, their agent can only see that same data. Security and governance rules travel with the person, not around a separate, harder-to-track service account.
Asked what leaders in other service industries should be thinking about, Hunnicut pointed back to the litmus test: audit your key systems for MCP or CLI access now, and if a vendor doesn't have it, ask whether it's on their roadmap before you commit further.
Neighbors' answer was more of a warning. However fast a leader thinks their organization is moving on AI, it isn't fast enough — the pace of change this year is unlike anything he's seen in 25-plus years in software. His biggest concern isn't reckless adoption; it's the opposite: organizations where governance debates consume all the oxygen in the room and crowd out the work of actually learning the technology. Governance still matters, but it can't be the only conversation happening, and leaders who aren't willing to get hands-on with these tools themselves won't recognize where the real opportunities are for their business.
"However fast you think you're going, you're not going fast enough."
— Derek Neighbors, CTO, Vixxo Facility Solutions
Watch the full conversation, The Litmus Test: Why MCP & CLI Integration Is Mandatory, from Deep Learning with PolyAI.