Day One Diaries

Collective Play

Early Learning Design & Fit-Out · August 2026

Collective Play design, supply and fit out early learning centres — and they are growing fast. The pressure that comes with that is a familiar one: more work, more people, and more of the business held in somebody's head.

The brief

Two people. First base. Before the team gets there.

Collective Play is a supply and fit-out business — room-by-room kits, inventory, purchasing, margin, and new-centre rollouts run as projects. Every one of those has detail that lives in a person rather than a system, and every one of them gets heavier as the business grows.

The room was deliberately small: Steven, the GM, and Lou, the owner. Just the two of them, on purpose — so the two people steering the business become the two most capable people in it on AI, rather than finding themselves led by whoever on the team got there first.

They were already on Claude, having moved across from Copilot. No Microsoft-house constraint, no data-security lock-in to work around — just a straight run at the best tool for the job.

How we designed the day

Fundamentals first — even when there's an itch to build.

Steven arrived sprinting. Four months into the GM seat, he turned up with his own five-layer AI framework already sketched out — which tells you everything about how he's approaching the role.

When someone arrives with that much momentum, the temptation is to skip ahead to "so where could we use this?" That question is a trap this early. Answer it before you understand the machinery and you end up with a shiny object every fortnight and nothing that compounds.

So we started underneath it.

What we did

Foundations first. Then straight onto real work.

1
Claude and Projects — properly
Why instructions matter, how files get uploaded, and what a Project actually holds and remembers — just as importantly, what it doesn't.
2
The ChatGPT comparison
Where the equivalent instructions, files and settings live in the other camp — so nobody ends up fluent in one vendor's vocabulary and stuck inside it.
3
Models, tokens and the machinery
The background stuff. Unglamorous, and the reason everything afterwards makes sense.
4
Recorded a real conversation
Steven and Lou spent five minutes talking through a role they're hiring for — an interior stylist — what they actually want in the person, and how the business recruits. Captured with a transcript tool.
5
Built their first Project
Chat, project instructions and files — with that conversation loaded in alongside the position description they already had written.

Then, without being asked: Claude flagged the gaps between the written position description and what Steven and Lou had actually said out loud. Nobody prompted it to check. That is the difference between a feature demo and an application — and it landed on their own work, in front of them.

The back half of the day

Plumbing for what he builds next.

With the fundamentals in, we spent the rest of the day designing — not building — the thing Steven wanted next: a single meetings "brain" covering his four new key meetings.

One Project, four foundation files: Sales, Purchasing, Individual, Whole Business. Four files inside one project rather than four separate projects. One brain, four facets.

The mechanics are where the real value sat:

Instructions versus files. Instructions are how it behaves. Files are what it knows. Folder-steering rules belong in instructions, never buried inside a document.
How a shared project actually works across a team. Instructions and files are shared; chats stay private; connectors don't travel between people. Which is precisely why foundation content has to live in project files rather than hanging off one person's account.
Getting work out of a chat and back into a file is a deliberate two-step. Ask for the complete updated file, download it, replace the old one. Nothing commits itself — and it will quietly summarise away your detail unless you tell it not to.

Designed, not built. That was the point. It's his to run.

In Steven's words

The foundation was the win.

Steven's own read on the day was that the value was the foundation — understanding what this is and what it's actually for. He was open about the pull to leap ahead to "where could we use this", and about deliberately holding himself back from it.

"Align the AI journey with our growth journey — but bring it back to the fundamentals first."

— Steven Hallett, General Manager, Collective Play

His plan, in his own framing: small trials, mini projects, build the fundamentals, then judge the bigger questions from a position of actually knowing. A word he kept reaching for through the afternoon was "ecosystem" — a mental model forming in real time.

Mark Calabro with Steven Hallett in the Collective Play warehouse

Mark and Steven, in among the racking at Collective Play.

The parking lot

Named, parked, and deliberately not started.

Not everything belongs in one day. Some of it we named and left exactly where it was:

The meetings brain — designed on the day, and his to build
A trial putting an account manager onto recording client conversations, with clear rules about what gets captured and how it gets used
The bigger structural questions — CRM, ERP, build versus buy — raised by Steven himself, and bigger than a single day
A quoting and pricing tool — the one I actively steered away from. A high-stakes estimator is the last thing you build, not the first.

Steven left with a method, not a deliverable. For a business growing at this pace, that is the outcome that matters — because the method is what keeps working after I've gone home.

Collective Play

Two people. First base. Before the team gets there.

Fundamentals first — even when there's an itch to build.

Foundations first. Then straight onto real work.

Claude and Projects — properly

The ChatGPT comparison

Models, tokens and the machinery

Recorded a real conversation

Built their first Project

Plumbing for what he builds next.

The foundation was the win.

Named, parked, and deliberately not started.

Want a day like this?