On Friday at Cube, we put our regular work aside for a day. No clearing out tickets, no wrapping up ongoing projects. Instead: a question, an idea, and a few hours to see if we could turn it into something that works. And it paid off quite nicely. The common thread? Many teams built something that didn’t exist yet. Sometimes because an existing solution just didn’t quite fit the way we want to work. Sometimes because we were curious to see if a new technology could make things easier for us. And sometimes simply because we thought: can’t we just build this ourselves?
Let’s just build it first.
One of the teams started with a fairly specific question: Can you also use AI for questions that don’t require a detailed answer at all? They tested System One by Jev AI, a model that doesn’t write but selects from predefined options. For example, to sort error messages, evaluate builds, or label reviews. The conclusion after a day of testing was exactly what you’d hope for from a hackathon: the technology works, but especially when the question is clear and well-defined. If another AI agent first has to figure out what the question actually is, a large part of the speed advantage is lost. So it’s not the solution to everything. But it is an interesting additional link between data and the large language models.
Another team tackled something we’d been struggling with for a while: our cars. No existing app worked exactly the way we wanted it to for mileage tracking. So they built one themselves. A device in the car detects when a trip starts and ends. Trips can then be added or edited via a web app. Even the technology was part of the experiment: the team deliberately chose something they don’t normally work with. With AI, that barrier turned out to be much lower. What would likely have remained just an idea without AI was actually up and running by the end of the day.
Even small things are worth trying.
Not everything we built was big. For example, we looked at Cap, an open-source alternative to the well-known spam protection for forms. No tracking, no user behavior data sent to Google, and relatively easy to host yourself. We’ll be running it alongside our current solution in the coming weeks to see how well it performs in practice.
We also built a way to manage a website using an AI assistant. It doesn’t just automatically adjust everything; instead, it operates within clear boundaries: the assistant can only do what the user is allowed to do, and everything remains a draft at first. During development, the AI reviews also revealed something unexpected: a few genuine security issues that would likely have gone unnoticed otherwise.
Meanwhile, a major cleanup took place behind the scenes. Years’ worth of reusable building blocks were scattered across an old system. These have now been consolidated in one place, right next to our code. The team had the AI handle part of the sorting work. A basic package was also created, including monitoring, logging, and roles and permissions, so that new projects don’t always have to start from scratch.
From separate systems to a single overview.
There was also a team that was especially fed up with searching. Because what’s the actual status of our projects? Which clients are active, what opportunities are out there, and how much work is coming up? The answer was out there somewhere, but scattered across different systems. So they built an internal app that brings all the information together. Not a new system to keep track of yet another system, but rather a single place that makes use of the data that’s already there. That’s perhaps typical of a hackathon: you don’t always have to come up with something completely new. Sometimes it’s already a win when something that was unnecessarily complicated suddenly becomes a lot simpler.
And then there was one team working on a question we’re hearing more and more often: if you can build a useful app with AI in an afternoon, where do you safely run it? They worked on a place where you can build apps and put them live right away, while keeping your data secure behind a layer we manage. And if an app proves to be useful, it can simply grow there.
Five minutes. In theory.
At the end of the day, each team had five minutes to show what they’d built. Five minutes, that is. For enthusiastic developers, that turned out to be about as realistic as a meeting that ends right on time. What was just an idea on a list that morning was suddenly running on a laptop by the afternoon. Sometimes the answer was a resounding “yes.” Sometimes an honest “not yet.” And sometimes, above all: interesting enough to explore further. That’s exactly why we hold days like this.
In between building, the wellness week was also in full swing: there was a workshop on sleep, and a healthy lunch was provided. We also celebrated Jasper’s five-year anniversary at Cube: party time!
It was a (healthy) day of building, testing, learning, experimenting, and, above all, seeing lots of things work that didn’t even exist that morning. Another great day.
What would you like to build?
Fortunately, you don’t have to organize a hackathon to explore an idea. Tell us what you’d like to do smarter, easier, or differently. We’d love to brainstorm with you, no strings attached.
Worth reading next...
Our hackathon recap: code, pizza, and music bingo.
A day of building without brakes. Our hackathon recap.
