When AI assistants became part of how construction teams work, ProjectTeam had a choice to make: build our own AI, wait, or connect to the tools customers already use. ProjectTeam’s VP of Product, Josh Witmer, led that decision and the development of our MCP connector. Continue reading for Josh’s perspective on how we got there, originally shared on LinkedIn.
For decades, building software meant building for people. Teams designed screens, forms and workflows, and measured success by how easily someone could click through them. That’s still part of the job, but over the past year I’ve started thinking more about a second kind of user.
More of our customers work with AI assistants now, and those assistants need to understand our system, too. So we changed how we build at ProjectTeam. When we design something new, we ask whether an AI agent could understand it and use it to help the person on the other side of the chat.
The first big result of that thinking is our MCP connector, which links ProjectTeam to the AI tools our customers already use, starting with ChatGPT, Claude, and Gemini. As far as we know, we’re the first capital program and construction management platform to offer this. I wanted to share how we got there.
Our customers run public capital programs: new buildings, roads, utilities, facility upgrades. All of their project records live in ProjectTeam, from RFIs and submittals to daily reports, inspections, schedules, budgets, and change orders.
But we kept seeing people do their thinking somewhere else. Someone would open ChatGPT to draft a memo or summarize a meeting, then manually copy project details over and paste the results back. It worked, sort of, but it was slow, easy to get wrong, and the AI never had the full picture.
We could build our own AI assistant inside ProjectTeam, wait a year or two to see where the market landed, or connect ProjectTeam to the tools people were already using.
Building our own didn’t make sense for a company our size. The big AI companies are spending billions on their models, and our strength has never been the model. It’s the project data and the workflows around it. Waiting was tempting, since a lot of our customers are still cautious about AI. But people were already using these tools at work, with or without us, and I’d rather they work from accurate project data than from whatever got pasted in.
So we went with MCP, the Model Context Protocol. It’s an open standard that lets an AI assistant securely read and work with data in another system. A few things made it the right fit for us.
ProjectTeam is where the official project record lives, so we’re the natural place for an AI assistant to get its information.
Our platform also brings owners, construction managers, designers, and contractors into one shared space with the same security controls, which is something most tools in our industry don’t do. The connector lets each of those groups use AI with that same trusted data.
And because MCP is an open standard, ProjectTeam can work alongside other tools that support it. Someone’s AI assistant can pull from ProjectTeam and their other systems in one conversation. I didn’t want AI to be something that only works inside our product.
Before any development started, we focused on 10 key use cases our customers needed help with. Things like turning rough field notes into a logged daily report, creating an RFI without filling out every field, or getting a quick read on where a whole program stands. That list became our scope, and we used it to judge whether each early version was working.
I worked side by side with our development leads to test those early builds on realistic project scenarios. Some things worked right away. Others took several rounds. At the same time, I had calls with customers to show them where we were and hear what they thought. Their feedback changed what we prioritized and how we talk about it.
Security was the first priority. ProjectTeam has always taken security seriously, and we maintain separate FedRAMP and GovRAMP environments for customers who need them. We weren’t going to cut corners for AI. Organizations can turn off AI connections completely if they’re not ready, because that should be their decision. When AI is connected, it follows the user’s existing permissions exactly, so it can only see and do what that person could already see and do in ProjectTeam.
The other concern we hear a lot is, “We don’t want our people to stop thinking.” I understand that, and it shaped how we approached the AI efforts. We’re trying to eliminate the repetitive work that fills a project manager’s day, like retyping notes into forms, digging for the latest status, or rebuilding the same report every week. The people running the project still make the decisions. They just get more time to make them.
Today, users can ask their AI assistant to:
I’ve shown this in YouTube videos, at conference sessions, and in a recent live webinar that drew about 175 registrations. The response has been better than I expected. Customers who have been with us for years are excited again, and several have told us they’re glad to see us move this early. Once people see it working on the kind of data they deal with every day, the conversation shifts quickly from “Is this safe?” to “How do we get started?”
We’re going to keep building on this based on how customers use it. Construction has a reputation for being slow to adopt new technology, but I don’t think it has to play out that way with AI.
If you manage capital programs and want to see it, schedule a demo with me. I’m happy to walk you through it.