In-Line Projects builds and delivers construction projects across Melbourne — and like a lot of firms this size, the front end of the business runs through the two people least able to spare the time. Estimating in particular sits with James, one of the two directors. When the person doing the estimating is also one of the people trying to grow the business, every hour in the weeds is an hour not spent on the bigger picture.
Free up the front end, estimating especially. The instinct with a brief like that is to charge straight at the biggest, gnarliest build.
I pushed the other way — a high-stakes estimator is the last thing you automate, not the first; get it wrong and you won't always know.
So we spent the afternoon getting both directors genuinely capable and building on their own work, with the estimating deep-build teed up for when they're ready to take it on.
We started with the machinery — Claude and Projects, how instructions shape behaviour, what a project holds and remembers, and how a simple recap file becomes something the project draws on every time. Then the habit that makes it stick: one thing a day, on real work, until reaching for AI stops being a decision and becomes reflex.
In-Line lives in Microsoft — SharePoint, Copilot, one source of truth, and a healthy preference for keeping it that way. So the answer was never "switch." It was: use Claude to think a problem through and work out what to do, then take that back into Copilot, where the business actually runs.
"Start with small bits — basic instructions that create a set of rules to guide what you're looking for."
"These micro habits — just doing one thing every day. I know it's there, but seeing the power of what it can do is the shift."
This is where it got real. Working through genuine business problems, we built two things and solved a third.
For a role they're hiring — fed in the org chart and a couple of existing PDs, and it came back with a strong draft from limited input. It even pulled a salary range from what it could find in their own system that landed close to the mark.
An agent that filters key project information and delivers a summary at 6am each Monday, so the week starts with a clear read on where every project sits. Still a little rough, and honest about why: that's the prompting and the inputs, not the idea. Checked against Daniel's real site visit that morning, it held up.
"From my site visit this morning, it's very, very accurate — which is pretty amazing."
James went after his own problem too — getting site photos off his phone without the usual clunky email chain — and left with three options, a recommendation, and a clear line on the time and inbox clutter it'll save.
Daniel put the real lesson better than I could:
"It can pretty much do anything — but you've got to give it the time to dedicate to it, which ultimately saves you time in the long run."
That, and the reason the directors — not the tools — stay in charge: AI is convincing, and when it gets something wrong, it rarely tells you. Keeping your hands on the wheel isn't caution for its own sake. It's the whole point.
Three and a half hours, start to finish, done by half four. Both directors left more confident, with working tools built on their own work and a method to keep going. The estimating deep-build — the thing the brief started with — is teed up as a scoped next step, taken on from a position of actually knowing how the tools work, rather than forced on day one.
Take the free AI assessment — a few quick questions, and I'll come back to you personally.
Take the Free AI Assessment →