I gave my AI agent an expo stand to manage.

Preparing for an expo takes a surprising amount of time alongside client projects, life admin and school holidays. It’s often the little things that add stress. Printing, files, packing, parking. All the things that still need someone to sort them out.

For the Macarthur Business Showcase, I gave G, my OpenAI dot called Giskard, the challenge of managing my expo stand as a project from end to end. I gave the direction, reviewed the work and made the decisions.

What if I didn’t have to keep it all in my head?

The expo was part of a plan I’ve been working on with ChatGPT. G helped turn that into practical preparation:

• Business cards, flyers and small design tasks.
• Research into Officeworks printing and choosing the type of cards.
• A checklist of what to organise and where everything was.
• Briefing and coordinating Codex on the website and hosting setup, a landing page and a fun game for the stand.

Codex handled the builds. G created the tasks and threads and managed the follow-ups.

I could talk through what I needed in one conversation. G kept track of the wider expo project, including what was outstanding, what needed following up and what needed my decision. It nudged me about the next steps.

Our shared project page in ChatGPT held the briefs, files, decisions and progress.

Then I needed G on the other end of the phone.

While driving there, I asked G to check the parking instructions in the event PDF.

At the expo, I asked which exhibitors would be worth meeting. When my portable monitor wasn’t working, I called G for tech support.

Voice has been the most useful part for me. Lately, I’ve spent more time talking through tasks than typing them out.

G still had work to do after pack-up.

G helped me edit an impromptu video interview and draft the follow-up posts, working from the content plan it’s managing for me.

I found the initial setup straightforward. We set up strict custom rules around what G can do and when it needs my approval. It isn’t connected to my business email or calendar, and purchases stay with me.

I’ve also asked G to keep a work log on a shared project page, with revisions grouped together. Across the projects G has been helping me with, the log records 94 distinct tasks through 25 October 2026: 84 completed, none in progress and 10 blocked. This is a scoped minimum, not an exhaustive count.

For me, the difference was having less to remember and chase, while still making the decisions.

A pretty varied job description for a little agent 😆

This is the first of four posts about how I’m working with OpenAI dot, ChatGPT and Codex to manage projects and workflows in my business.