From My Computer to the Team: Our First Shared Project with the Starter Kit

Over the past weeks and months, I have devoted much of my time to developing the second version of our internal Starter Kit. Before our team-building weekend, I was finishing its release candidate: a version we wanted to test by working together on a concrete project.

By the end of the weekend, I saw something I had been looking forward to for a long time: projects built on this foundation were running on my colleagues’ computers, complete with features they had created during our time together.

After so much time spent designing, developing, and testing on my own machine, that was a special feeling. The shared foundation was helping other people, and I could see the results of their work firsthand.

From design to our first shared experience

In my previous article, I described my approach to an internal CMS and a shared application foundation built with Laravel. I covered the separation into a core platform, plugins, themes, and project customizations, along with the challenges of designing updates and developer interfaces.

If you are unfamiliar with the term, a starter kit is a prepared foundation for building applications. It brings together commonly needed components so developers do not have to create them again for every new project.

The current design also draws on our experience with the first version. It gave us useful lessons that we could incorporate into the new design. The second version builds on what we have already learned.

The team-building weekend gave us an opportunity to try this foundation together and see how it worked for people who had not designed it themselves.

Eight people and a shared goal

There were eight of us, with very different levels of programming experience. Only three of us are programmers with a deeper background in software development. The other five do not program professionally: some are familiar with it, while others had practically no experience before this weekend.

We began by discussing how we use AI and incorporate it into our everyday work. I then presented the Starter Kit we are preparing to make development easier across client projects.

I will not go into my colleagues’ specific methods or the content of our internal discussions here. They are part of our company’s know-how. The kit itself and the details of the application we worked on also remain private.

What I can share is that it was a marketing application. Its purpose is to simplify a company workflow involving several steps, making it possible to manage the process in just a few clicks.

We had a concrete brief and several questions to explore: could the technologies and connections to other services we had selected deliver the intended result? And how would the prepared foundation help us get there?

What we actually validated over the weekend

The result was a pilot project to validate the concept, often called a proof of concept. We tried out the Starter Kit as well as the technologies and integrations needed for the application we had in mind.

We divided the work into smaller functional areas and assigned them to individuals or small teams. Each had a defined area and an agreed set of features to work on.

At this stage, being able to work independently was important. Although the steps are intended to connect in the final application, we could explore their feasibility separately. Connecting every part into a single system was not a prerequisite for this initial validation.

At the end, we therefore focused on the results of the individual parts and what they told us about the feasibility of the overall idea. Testing the complete workflow in a connected application is a further step.

A shared foundation allows individual parts to be developed independently. Bringing them together creates the complete solution.

AI agents as part of the development foundation

From the outset, I have been preparing the Starter Kit for a workflow in which development takes place through AI agents. I refer to this as AI-first development. Here, AI agents are artificial intelligence tools that help write and modify application code based on the tasks they are given.

In practice, an agent receives project context alongside its task: instructions, conventions, descriptions of the available capabilities, and skills prepared for specific tasks. This gives it a basis for deciding where a new feature belongs and how to integrate it into the application.

I have put a lot of effort into preparing that context. Knowledge about how the shared foundation works needs to be available outside its author’s head. Otherwise, colleagues would have to keep coming back to me to ask how I intended a particular part to work.

We used this approach while developing individual features over the weekend. For me, it was an opportunity to see what others could create with the foundation I had prepared.

Getting started: the AI installer and a small permissions fix

The only hurdle when assembling a project from several private GitHub repositories was resolved with a small adjustment to permissions on GitHub. It was not a bug in the Starter Kit. During this pilot, we needed to verify access to its individual components from our colleagues’ accounts as well.

The kit was already prepared to help people get a project running. It includes an AI installer that assists with installation and setup on a work computer. It accommodates macOS on MacBooks as well as Windows and Linux, giving colleagues support during the initial setup regardless of their operating system.

When I spend a long time preparing a shared foundation on my own computer, the necessary permissions and some of the setup are already in place. Using it on other machines let us validate this part of getting started together. The installer helps prepare the environment, while access to private repositories still depends on correctly configured permissions.

The highlight: seeing results on my colleagues’ computers

What made me happiest was the final demonstration of what the individuals and small teams had created during the weekend. Broadly speaking, everyone managed to make their assigned parts work, including colleagues with no previous programming experience. Given their different starting points, those results meant a lot to me.

Within this pilot, the prepared foundation and AI agents helped people who do not normally develop software contribute to building an application.

I saw projects built on our shared foundation running on my colleagues’ computers, with tangible results that included more complex logic. After weeks and months of working on the second version of the kit, that felt like real progress.

Based on my previous experience, I would have expected some of these features to take several days, perhaps even weeks, to develop without AI agents. Over the weekend, I saw their first working versions.

That is my personal experience from this pilot, not a measurement of development speed. The team’s focus, the division of tasks, the prepared foundation, and the use of AI all contributed to the outcome. Even so, the amount and complexity of what people created left a strong impression on me.

What I took away from the weekend

The weekend allowed us to explore the feasibility of the planned application and try out the Starter Kit with eight people working together.

For me personally, it was also an important milestone in developing the second version. Lessons from the previous kit, months of preparation, and decisions about its design began to translate into other people’s work.

In the previous article, I wrote about what I expect from a shared foundation. This team-building weekend gave us our first shared experience of putting those expectations to the test on a concrete project.

Above all, I came away happy that the foundation I had spent so long preparing was now helping others build their own solutions.