)
)
TL;DR
We ran pretty much the same experiment ten months apart and accidentally got a very good snapshot of how much Cube had changed. After six months of our Factory Rebuild, Hackathon Two went way beyond quick prototypes, with teams building things that could genuinely find their way into how we work. We also rinsed our AI tokens again, so there’s comfort in some things never changing.
For the uninitiated, our hack day is basically one day where everyone downs normal tools, picks a real problem, and sees how far they can take it.
Our second AI hackathon was a whole different beast
Cast your mind back to the days of yore: November 2025, to be exact. We ran our first AI Automation Hackathon at 3sided Cube. We gave the whole agency a day away from client work, armed ourselves with Replit and picked a bunch of those annoying, fiddly, why-are-we-still-doing-this-by-hand problems to see what AI could do about them.
The Cubies walked away with eight (kinda) working prototypes, and we thought we’d absolutely crushed it. To be fair to past us, it was pretty bloody cool. People who normally wouldn’t touch code with a ten-foot bargepole were suddenly getting ideas onto screens. And the Developers had the freedom to go buckwild. We built scheduling tools, accessibility experiments, analytics automations, and one creepy AF attempt at a 3D human which still haunts our nightmares.
We wrote a whole blog about the day and another one about what happened when we put Replit through its paces, because that little old hack day genuinely blew our minds.
There was one fairly important caveat though. If you weren’t technical, you generally needed to buddy up with someone who was. Replit massively shortened the distance between an idea and something you could actually play with, but when things got complicated there was still a point where you swivelled your chair towards a developer and begged for a software-buildin’ saviour.
That was ten months ago. AKA a literal lifetime in AI.
)
So when Hackathon Round Two came around in September, we expected it to be kind of par for the course. What we didn’t quite expect was for the whole thing to feel like it had been flipped turned upside down (in the best way).
You might be asking yourself, “Self, what could possibly have happened betwixt hackings to make them that different?!”
Well… buckle up, buttercup.
Hackathon Round Two: shit got real
The brief was similar to the first one: take something from our actual workflows and see what we could do with it. The Cubies gathered into their teams, rubbed elbows, scoffed pastries, and got to nerding.
By the end of the day, our commercial team had built a persona matching tool. Two Project Managers had got client project-plan generation down to around 8–10 minutes. Our tech teams were deep into automated QA, app scaffolding, accessibility triage and 24/7 support investigation.
And that’s where the comparison got interesting.
Last year, getting something kinda working often felt like the win. This time, teams were already talking about what they’d keep building, how it would connect to our existing systems, and where it could slot into the way we work once Monday rolled around.
)
The gap had shifted at both ends too. Non-technical Cubes could take their own ideas much further without technical help, while our developers could use the day to get into much meatier problems than “can we make a prototype?” Nobody suddenly became interchangeable. We just got to spend the day using our individual expertise to look at how we’d always done something, and dream up an entirely new way to do it… with an even better outcome.
Want to see exactly what all ten teams nerded out on? Get the full rundown here.
You may be asking yourself how the hell was any of that possible? And what AI devil did we sell our souls to?
No magic wand ✨
Just six months of completely changing how a 17-year-old software agency operates.
You may have heard us banging on about our Factory Rebuild. The mentionitis has been off the charts for the last few months, but not without reason.
Earlier this year, our Fearless Leader Duncan came in sideways with some classic CEO energy. Cube had been building software for 17 years, which comes with 17 years of processes, habits, and “this is how we do it”. Duncan decided bolting AI onto the edges of all that wasn’t ambitious enough.
We were going AI-AF.
Cue Regina George: get in the car, loser, we’re rebuilding the company.
)
We’ve documented the gory details across our Factory Rebuild series, because we’ve had a bit of mentionitis about taking apart how we’d worked for nearly two decades and putting it back together again. But we pinky promise to not get into it, AGAIN.
Here’s a whistlestop tour of the bits that changed the rules for Hack Day:
🚀 Everyone had somewhere to build. Our own infrastructure and VPS meant you could get an idea running and hosted without every experiment beginning with finding somebody who knew how to set everything up.
🤖 AI had context. Hive started bringing together the knowledge about Cube, our work and our projects that normally lives scattered across tools, documents and human brains, so we weren’t constantly starting every conversation with a blank box.
🧑💻 People had spent months actually working this way. Project Managers were building their own tools. Designers were getting closer to implementation. Marketing had somehow ended up making it rain pull requests in GitHub. Meanwhile, our developers were using AI deeper in the actual software-delivery process, from project scaffolding and testing to support investigation.
💚 Cube was the guinea pig. We experimented on our own workflows before confidently telling clients how much AI could transform theirs. There have been wins, cock-ups, expensive lessons, and at least one COO who may still be rocking gently in a corner. (Thoughts and prayers to Sophie.)
Those six months of actually working this way mattered just as much as the tech. People could take an idea further themselves, play with it, realise bits were crap, change them and try something else without the whole experiment having to pass through somebody else’s hands first.
The specialist expertise hasn’t become less important either. If anything, it gets used in a better place. Developers can spend more time making the hard technical decisions rather than being the only route between somebody else’s idea and its first rough incarnation.
Hackathon Two made that difference damn near impossible to miss.
What did getting our hack on actually look like?
Take our commercial team.
They used the day to see how much more useful a CRM could be if it did some of the digging before a human ever opened a contact record. Their prototype pulled together things like website behaviour, company information and lead scoring, then used that context to understand who someone might be, how they fit our buyer personas and how we should actually talk to them.
By the end of the day, there was a working interface to poke around in rather than a Figma board describing something we’d quite like somebody else to build one day. It ran our Sophie through the persona matching and concluded that she was both a “level head” and “goddess divine”, while clocking her love of Cher.
Clearly the technology has reached full maturity.
)
Then there were Tom and Lou, who attacked the eternal Project Manager problem of creating a lovely, in-depth project plan only for the schedule to change fourteen seconds later.
They combined work they’d already been tinkering with and got a client-facing plan generated from live schedule data in around 8–10 minutes, complete with editable milestones and its own URL. More importantly, by the end of Hack Day the next conversation wasn’t “could this work?” It was how to automate the updates and pull it properly into the project workflow.
Our accessibility team went HAM on hacking too. They took a real external audit with 56 findings and built an AI first-pass that could assess the evidence, check it against Web Content Accessibility Guidelines, and decide whether something looked valid, invalid, or still needed a human. When they tested it against the existing audit, it independently disagreed with 33 of the original findings.
That isn’t a “humans are obsolete” use case. It’s actually the opposite. If AI can chew through some of the repetitive first-pass work, our accessibility experts get more time for the nuanced stuff that needs actual expert TLC.
)
And for a properly techy example, one team tackled what happens after code has been merged.
They built an automated QA workflow that could grab the build, pull context from Jira, Figma, the codebase, and TestRail, run the tests on a simulator, gather screenshots and video, and either move the ticket through if everything passed or kick it back open with details of what had gone wrong.
That’s a very different Hack Day output from knocking together a nice-looking prototype.
There were six more builds covering everything from app scaffolding to app-review analysis, DesignKit automation, 24/7 support investigation and, yes, Jake teaching a mapping tool how to automatically scatter tents around Festival campsites.
No one had “AI-powered tent scattering” on the strategic roadmap, but here we are.
Rather than turn this into the extended Lord of the Rings edition almighty read for you, we’ve put the full rundown of all the things we worked on over here!
Fancy running your own hack day? Here’s our top tips!
After two of these things, we’re basically seasoned professionals (please do not fact-check that).
Start with real annoyances. The teams that got furthest weren’t inventing hackathon ideas on the morning. They brought problems they already understood and actually wanted to fix.
Give people the playground before the day starts. Access, infrastructure, context, and the right tools matter. Six hours disappears very quickly if hour one is spent asking for permissions.
Don’t force everyone into the same shape. Some problems need a proper cross-functional team. Some are better owned by the person who knows that problem inside out, with experts dipping in when needed.
Decide what happens on Monday. A clever demo that dies when the pizza boxes go in the bin is just a clever demo. Give the useful stuff an owner and a next step.
Buy more tokens than you think you need. BUY ALL THE TOKENS. We started with a sensible plan about which models everyone should use and when. By late afternoon, Duncan was hiding in a corner while various Cubes tracked him down asking for more. He’ll be paying off our token debt for the foreseeable future.
)
So what actually changed?
The easiest answer is that the starting line moved.
Last year, giving non-technical people the ability to prototype at all felt transformative. This year, people across disciplines could take something they understood deeply from their own working lives and get much further into solving it themselves, while our technical teams could go deeper into the complicated stuff.
That meant better conversations when specialists did get involved, more room for experimentation and a much higher chance that whatever got built on Hack Day would still be alive the following week.
For all the noise around being “AI-native”, that’s probably the clearest proof we’ve had that our Factory Rebuild is actually changing how Cube works.
Where will we be by Hackathon Three? God knows. Give us another few months, and we’ll either be solving half the world’s problems or Duncan will have put the token budget behind retinal scanning.
Place your bets now!
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 is an AI hackathon?
An AI (artificial intelligence) hackathon is a day where a team puts normal work down, picks a real problem from their own workflows and uses AI tools to see how far they can get at fixing it. At 3 Sided Cube, the whole agency steps away from client work for the day. Pastries are involved. Heavily.
How do you run a successful AI hackathon?
Start with real annoyances people already understand. Sort out access, tools and infrastructure before the day, because six hours disappears fast. Decide who owns the useful stuff come Monday. And buy more tokens than you think you need. Then buy more. Ask Duncan.
How was 3 Sided Cube's second AI hackathon different from the first?
The starting line moved. At our first hackathon in November 2025, getting a prototype kinda working was the win, and non-technical Cubes usually needed a developer buddy. Ten months later, non-technical teams took their own ideas much further solo. Developers went after meatier problems like automated QA (quality assurance). And the chat at the end of the day was about what we'd keep building on Monday.
What is 3 Sided Cube's Factory Rebuild?
Our Factory Rebuild is how 3 Sided Cube is going AI-native. We're rebuilding 17 years of processes and habits rather than bolting AI onto the edges. It gave everyone somewhere to build and host ideas. It gave AI real context about our work through Hive, our knowledge base.
Published on 29 September 2026, last updated on 29 September 2026