Four Days to 30 Minutes: Building an AI Assistant for Emergency Management Logistics Canada

TL;DR

Scott Cameron runs Canada’s only shared online emergency management logistics networking platform. When communities face disasters — floods, wildfires, or mass evacuations — local governments and community organizations rely on coordinated systems to identify resources and connect support quickly.

He’s built something genuinely important.

Scott is not a developer. He runs this platform as a founder-operator, which means the website is his to maintain. And like a lot of mission-driven founders, he’d been making it work with the tools available to him—including AI.

That work matters. Scott’s time matters.

Scott was using AI to make updates to his website, which was a great step! Scott was getting it done.
But As a non-technical founder, he was still four days wrestling manually copying code into ChatGPT, waiting for suggestions, copying it back, there was room to improve.

He was the physical bridge between an intelligent tool and his own website. Every edit required him to act as courier, translator, and typist all at once. The AI had plenty to say. It just had no way to reach anything. The Other challenge? Brilliant Directories was not built for AI. So we had to find a way to connect Claude to the website.

And find, we did! the guiding question was: what does this need to feel like for someone who just wants their website to work?

Scott’s now set up in Claude and built a direct connection with the EML Canada website—so Claude can now read his pages, see what he’s working on, and make edits directly, with Scott’s explicit confirmation before anything goes live.

A challenge that had (reasonably) consumed four days the previous week was resolved in 30 minutes.

Positive Outcomes

• 4 days of work compressed to 30 minutes: The specific challenge Scott had been grinding through was resolved the first time he used the new setup
• Claude reads the site directly: No more copy-pasting code or content into a chat window—Claude has eyes on the actual pages
• Edits happen with confirmation, not chaos: Claude can make changes directly, but nothing goes live without Scott saying so—he’s in control, not out of the loop
• Plain English, not developer jargon: Claude responds like a trusted colleague who happens to know the site inside out, not like a technical tool Scott has to interpret
• Time returned to the mission: Hours that were disappearing into website maintenance are now available for the work Scott is actually here to do—building infrastructure for emergency preparedness across Canada

The Full Story

Scott built Emergency Management Logistics Canada (EMLCanada) to strengthen those connections. The platform links municipalities, community organizations, and emergency management professionals so they can identify local assets, build relationships, and coordinate more effectively when emergencies arise. Today, dozens of municipalities across Alberta — and a growing number across Canada — use the platform to support preparedness, response, and recovery efforts.

EMLCanada has also participated in collaborative pilot initiatives with national organizations, helping explore how shared logistics networks can improve coordination when disasters occur.

The Challenge

The problem wasn’t that ChatGPT was unhelpful. It was that using it required Scott to do everything manually. He’d identify a challenge on his site. Then he’d dig through his files to find the relevant code. Then he’d copy it into a chat window. Then he’d read the suggestions back. Then he’d copy those changes into his site. Then something would be slightly off, and he’d start over.

There’s a word for that kind of arrangement: assistant. Except Scott was the one doing the assisting.

One week before we started working together, a single website challenge had taken four days to work through. Four days of context-switching, of copy-paste loops, of trying to hold the thread of a technical problem while also running an organization.

That’s not a workflow problem. That’s a structural mismatch—between a powerful tool and a person who shouldn’t have to act as its hands.

Our Approach

Most people think about AI as a thing you talk to. You ask it something, it tells you something, you go do it.

That model works fine for some things. But when your work lives on a website—and the AI has no access to that website—every interaction requires you to bridge the gap manually. You become the connector. And connectors get worn down.

The question we asked wasn’t “how do we help Scott use AI better?” It was “how do we get the AI into the room where the work actually happens?”

Scott’s website runs on Brilliant Directories. Through research, we found a way to connect Claude directly to it. That was the opening.

We built a direct connection between Claude and Scott’s site. Not a workaround. Not a smarter copy-paste routine. An actual connection—so Claude can read the pages Scott is working on, see the content as it exists, and make changes directly when asked.

We made two deliberate choices in how we set this up:

First, safety over speed. Claude can read and edit, but it cannot delete anything. Ever. No matter what. That’s not a limitation we might revisit—it’s a design principle. Scott should never have to worry that an AI assistant, however well-intentioned, removes something it shouldn’t.

Second, confirmation before action. Claude can draft and propose changes, but nothing goes live until Scott explicitly says so. This keeps Scott in control of his own platform. The AI is doing the work; Scott is still the decision-maker.

And because Scott is not a developer—and shouldn’t have to be—Claude was set up to communicate in plain English. Not code snippets. Not technical explanations. Guidance that sounds like a trusted colleague who happens to know the site inside out.

Implementation Process

The setup came together in a focused working session. We found the connection point for Scott’s platform, built the integration, and configured Claude with enough context about his site to be genuinely useful rather than generically capable.

We also worked through the communication layer—the part that determines whether an AI feels like a colleague or a search engine. Claude was given context about how Scott’s site is structured, what he’s trying to accomplish, and how to respond in a way that moves work forward rather than creating more questions.

Then Scott tested it.

The challenge he’d been working through the prior week—four days of back-and-forth—went back on the board. With the direct connection in place, Claude could read the relevant pages without being fed them manually. It could see the context. It could make a targeted suggestion. Scott confirmed. It was done.

Thirty minutes.

There was no magic in that number. There was just the difference between a tool that has access to your work and one that doesn’t.

How did it turn out?

“The challenge that I’d been working through with ChatGPT that took four days last week were resolved using this new system in 30 minutes! While there are 100 things I’d like for it to do on my BD site, I have to remain focused on the big tasks and business strategy. Lots of time to tweak and experiment later on!”

 — Scott Cameron, Founder, Emergency Management Logistics Canada

The Bigger Shift:

That last line is the one worth sitting with: “I have to remain focused on the big tasks and business strategy.”

Before, website maintenance was pulling Scott out of that focus. Not because the problems were insurmountable—they weren’t—but because the process of solving them was slow, manual, and exhausting in a way that had nothing to do with the problems themselves.

Now, the AI handles the website work. Scott handles the mission.

That’s not a small thing for an organization like EML Canada. The work Scott is doing—building Canada’s infrastructure for emergency coordination—requires his attention at the strategic level. It requires him thinking about how organizations connect, how communities prepare, how response improves over time.

That time is back. And it belongs to something that actually needs it.

Why This Worked

The breakthrough here wasn’t a better prompt or a smarter AI. It was removing the gap between the AI and the work.

When you have to manually carry information back and forth between a tool and your actual environment, you’re not using AI—you’re transcribing it. The cognitive load of that process adds up fast. And it compounds: every failed attempt requires another loop, more copying, more translating.

Giving Claude a direct line to Scott’s website didn’t make the AI smarter. It made the setup honest. Claude could finally see what it was working with, instead of relying on Scott to describe it.

The safety layer—no deletions, explicit confirmation before any change goes live—meant Scott could give Claude access without giving up control. That’s the right balance. A founder shouldn’t have to choose between efficiency and oversight.

And the plain-English communication layer meant the tool actually fit the person using it. Scott doesn’t need to become more technical to benefit from this. The AI adapts to him—not the other way around.

The Takeaway

If you’re using AI but you’re the one doing all the moving—finding the files, copying the content, carrying the output back and forth—then you’re not getting the leverage you think you are.

The question isn’t whether AI is useful. It’s whether it’s connected to where your work actually lives.

When it is, four days becomes thirty minutes. And the work that actually matters gets your attention back.

Scott never had to become a different person to use this. He didn’t need to learn new vocabulary, develop technical intuition, or manage a tool that constantly reminded him of what he didn’t know. The system was designed to carry that weight so he wouldn’t have to. That’s not a feature. That’s the whole point.