Dots:
Day One.

We spent our first day with dots exploring a simple question: how might an AI assistant help carry a project from an idea into useful, ongoing work?

We tried a few things, built a few things, and learned where we need to be more deliberate. Before getting into the experiments, here’s a little context for anyone meeting dots for the first time.

What are dots?

Dots are AI agents from OpenAI in ChatGPT. Think of a dot as an assistant you can give an ongoing goal. It can continue working between conversations, use apps you choose to connect, and bring back results for you to review.

That might mean researching a question, preparing a document, helping organize a project, or building software. You provide the direction and decide what it can do. Your dot asks for decisions that need your judgment, and you can review its work or change direction along the way.

OpenAI’s overview describes the capabilities and current access details. For us, the interesting question was how those capabilities might fit into the everyday work of the Wasatch Lab.

Exploring with a new site

We started by setting up a new site as a place to explore day-to-day workflows that could be monitored or automated, with connections to outside sources. It gave us something concrete to try: bring a few pieces together, see where they were useful, and notice what still needed attention.

We also worked on plugins and a clearer place to organize them. Plugins give an assistant a way to use particular tools or services. Our goal was to explore how those connections could support repeatable tasks, so each experiment could teach us something useful for the next one.

Along the way, we spent time on the smaller details around projects: records, preview artwork, and branding. Those were good opportunities to test how well the assistant could follow a task through its different parts and help us keep track of the result.

What we learned

The connections need as much thought as the individual pieces. A site, a plugin, and an assistant may each be useful, but we still have to work out what information belongs where and what should happen next.

One lesson was to keep viewing and editing permissions separate. What a visitor can see is one question. What an assistant has permission to work on is another. We want to explore useful connections while keeping those boundaries clear and leaving important changes for human approval.

We also learned to be careful about what we call finished. Setting something up is a starting point. Trying it with real work, checking the result, and adjusting what didn’t work are part of the same process. Day one gave us a chance to begin that cycle.

Where we want to take it

As we explore further, the goal is to make these pieces work together in ways that are genuinely helpful: less time gathering the same context, clearer handoffs between tasks, and tools we can return to and improve.

There are plenty of questions left. Which connections will prove useful over time? Where does a simple workflow make more sense than another feature? How do we make the next step clearer without adding more things to manage?

Day one was great. We learned some things, built some things, and came away excited for what’s ahead. We’ll keep exploring, keep testing, and let what we learn shape the next version.