)
)
Ten teams. Six hours. One stretch of a challenge to take an AI experiment or idea and see how far we could take it in a day.
For our 2026 AI Hackathon, we spent one day off from client work to tackle our own workflows. We made things. We broke things. All of us ran out of tokens at some stage. Most importantly, we came away with a bunch of ideas that could actually make our working lives a whole lot better.
We’ve already shared the what, how, why, and what changed since last year, now it’s time to talk about the juicy details…
The tools we actually built!
1. Automated testing and design reviews
Team 1 looked at one of the most time intensive parts of building digital products, and that is acceptance testing, design reviews, and regression checks.
The idea was to automate more of the initial tests, meaning the team can keep an eye on the automated side while being able to spend that all-important Cube brainpower on the tests and reviews that truly need it.
And the live demo did exactly that! Team 1 dropped a ticket into a new "AI review" column in Jira and the agent took it from there. If anything fails, it bounces back to open with the bug details and a little robot emoji, so you know it was automated and exactly why it was sent back.
Admittedly, some of the Cubes are still warming up to the new column but the robot emoji is doing its very best to win them over.
2. Personal ideas, shared expertise
Because sharing is caring, Team 2 broke the mold and instead brought together individual problems they could work on side by side. They ran it somewhat like a think tank, swapping feedback, sense-checking each other's ideas, and lending the much needed bit of emotional support when Claude refused to behave.
Personal sticking points for reporting, handovers, and all that admin jazz, has now seen automations come to life!
While we can’t get into all the details on these because they are so personal, you can take our word for it when we say their approach worked! We’ve now got more personal automations in the mix than you can shake a stick at, which would not have been possible if Team 2 hadn’t banded together.
)
3. Customer Persona
Knowledge is power, right? So our Marketing and Client Success team came together to build a tool that does the digging on a new contact before a human even opens the contact record.
They also had another challenge on their hands too, to build the prototype without the help of a single developer.
By the end of the Hackathon, their prototype worked! It pulled together data like website behaviour and company information to understand a contact’s buyer persona as well as how the commercial team should be talking to them. And don’t worry, we’re still keeping all the actual outreach 100% Cube!
Team 3 got way further than ‘non-technical’ usually allows, mostly thanks to our 3 Sided Cube VPS set up, which meant they had a working interface to play around with.
4. Project plans without the faff
Two of our Project Managers tackled something they know all too well… manually editing project plans because one tweak creates multiple knock-on effects to the project and schedule.
The idea was to create a project plan hosted on its own domain that pulls live data from our schedule, so whenever a change is made, it automatically updates the project plans for Cube and clients! Even on the client side. This saves a massive amount of admin time, with much fewer project plans called FINAL_v7_REALLY_FINAL. Everyone wins!
By demo time, Team 4 had a plan going from schedule to its own shareable URL in around 8 to 10 minutes.
)
5. Accessibility triage
Team 5 went all in on accessibility. Some projects have huge amounts of accessibility tickets, and rightly so! But this means the team has to go in and manually check every single one. From the quick things to the complex multi-steps.
The aim was to triage these tickets by automatically assessing each one and then giving it a status of either valid, invalid, or needing human verification. Most importantly, the output needed to include the reasoning, supporting evidence, platform context, recommended action, and also flag if it couldn't confidently work out anything. That’s a whole lotta info!
By the end of the day, Team 5 had a prototype that was surprisingly close to that ideal. Now, the next step is to refine the tool even further so it’s more accurate before we start actually using it.
6. Making our DesignKit easier to scale
‘What in the world is ‘DesignKit’?’ I hear you ask! To put it simply, it’s an internal app that the team uses to see what different design components exist for a project, where they should be used, and how all the designs fit together.
It’s quite literally the ‘design kit’ for a project.
The challenge for Team 6 was making it easier to use. So, they worked on UX improvements, clearer versioning and update tracking, comparisons between Figma changes and live apps, plus a better migration and build process with a quality checker.
In other words, much less hunting around and much more knowing what's what.The live demo was a real Cube crowd-pleaser! DesignKit had a searchable, filterable component list, but the real magic happened behind the scenes. Mark a design as ready for development in Figma and Claude automatically opens a draft pull request, with the designer set as reviewer. Then, once code is merged, a GitHub action scans our projects to show exactly where each component is being used.
)
7. Meet Scaffoldy
Team 7 built Scaffoldy. Not just a cute name, it’s actually a set of Claude skills bundled into a plugin.
The idea was to help non-technical people create high quality technical outputs. You can give it some requirements and it helps you choose the right building blocks, explains why, sets up a GitHub repository and hosting, and also adds guardrails for what happens next.
Pretty damn handy when you’re an AI-native agency with non-technical people building their own tools!
The demo on screen summed it up perfectly: five questions in, one repo out. Scaffoldy even spotted that one of the templates it was meant to use wasn't actually finished, flagged it as a draft and refused to build on it. Told you it came with guardrails!
8. Better context for 24/7 support
When a 24/7 alert fires, the on-call developer has to gather context from a bunch of different places before they can decide what to do. And sometimes it happens smack bang in the middle of the night!
It’s really not ideal when you're half asleep, you know something is broken, and the likelihood is that it needs an urgent fix.
So, Team 8 looked at automating that first step by creating an incident briefing. It contains the observed impact, timeline, likely cause, supporting evidence, missing information, and a recommended action. The goal was not to replace the person on call. Instead, all that context gives them a starting point to work from, meaning quicker fixes. Cool right?!
The live demo worked, but with one catch. Testing it properly needs real incidents. Which, for once, is the one time we're hoping for a few. In the meantime, we know the groundwork checks out.
)
9. Maps that understand what a map is
Believe it or not, some maps don’t actually know that they’re maps! Existing tools are pretty good at drawing simple shapes, but they're way less good at knowing what the shape you're drawing actually represents.
On the first pass, you end up with various shapes that are meant to be a campsite, a path, and a stage, with no way to really distinguish them. That means a lot of manual work then goes into making them useful.
The prototype Team 9 worked on takes a different approach by having that understanding built in! In the demo, we saw areas drawn and reshaped on the fly, corners rounded off, shapes being cut out of one another, edges insetted, and then everything was sorted into tidy layers. The final party trick was one click that scattered tents across a camping area, all ready to export straight into Mapbox.
What seem like simple tweaks are turning a job that takes days into something that could take a couple of hours.
10. Competitive research, at super speed
Last but not least, Team 10 set their sights on New Business. Competitive research is part of every pitch and discovery phase we do, from App Store screenshots and ratings to feature comparisons and strategic opportunities.
All of that, understandably, takes time.
But what if it didn’t have to? The team already knew Claude Code could help, but the existing process lived in a zip file with setup instructions. So, not ideal if you aren’t familiar with code or Terminal.
Team 10’s solution was to take the zip file and build a prototype that's much easier to use. It’s now a branded HTML page that pulls in App Store data, screenshots, benchmark analysis, and more.
And their demo worked! A report still takes around 25 to 30 minutes, and results can vary a little between runs, so the next step is making it faster and more consistent. Either way, that's still a whole lot quicker than doing it by hand!
)
Worth every token
So, was our hackathon worth it? Abso-flipping-loutely!
Giving every single Cube the permission to try something, learn quickly, and see what's possible left a real buzzing energy that we wouldn’t change for the world.
While we do love having one day dedicated to coming together to hack away at problems, this kind of experimentation is becoming increasingly part of how we work every single day at 3 Sided Cube. And that’s across both technical and non-technical teams.
So, if this sounds like your cup of software development flavoured tea, come shout us a holla!
Disclaimer: We used AI in the process of writing this blog, but a human was still at the centre! We'd be poor advocates for going AI native if we didn't use AI to help write about it. The story is ours and every line got the human pass 💚
FAQs
What AI tools did 3 Sided Cube build at its 2026 hackathon?
At our 2026 AI hackathon, ten teams built ten prototypes pointed at our own workflows: automated testing, a think tank of personal projects, AI for our commercial team, self-updating project plans, accessibility triage, DesignKit improvements, the Scaffoldy plugin, 24/7 support briefings, a smarter map tool, and instant competitive research.
Can 3 Sided Cube build custom AI tools for my organisation?
Yes, and that's rather the point. Our teams built every one of these working prototypes for ourselves in a single day, so imagine what a proper project gets. Give us a Holla!
How long does it take to build an AI tool at a hackathon?
At 3 Sided Cube’s 2026 hackathon, it took one day. Every team went from idea to working prototype between breakfast and the afternoon demos. Some got further than others, which is exactly how hackathons tend to go.
Published on 29 September 2026, last updated on 29 September 2026