)
)
A Day of Building, Learning, and Levelling Up Together
As an AI-first agency, using AI isn’t something we just dabble with for special occasions; it’s part of the way we think, plan, design, and build. It sits quietly inside our workflows, speeding up the boring bits and helping us get to the good ideas faster.
So at some point we basically said to ourselves:
“Self…Cubies have 99 problems, and AI can solve all of them. Let’s carve out an entire day to experiment, fail (I mean learn) fast and up skill on new tools.”
A full day to pause the day-to-day hustle, look at the internal pain points, and actually do something about them with some handy dandy AI automation as our guide.
So we set the day, grabbed our laptops, and got ready to get stuck in…
Let The Hackathon Commence!
By 9am, Cube HQ had that familiar energy: lots o’ coffee, quick huddles, laptops galore, and a FigJam board full of problems we’d all complained about at some point.
The board looked like a very honest snapshot of life at Cube:
PMs manually scheduling 10–12 recurring meetings every project
Designers comparing app screens to Figma by eye
QA rewriting incomplete bug tickets
Ops battling spreadsheets
Teams trying to locate internal knowledge “from that one message someone definitely sent once”
HR juggling onboarding admin
Watch this to get a sense of the feeling in the air:
One team was already elbow-deep in 3D modelling before most of us had even fired up our machines, creating what they later lovingly called:
“the monster that we created.”
)
Another team had cracked open Widget Builder and announced confidently that they could probably automate scheduling by lunchtime.
And Sam, representing every caffeine-dependent human in the room, summed up the 10am brain fog perfectly:)
The vibe was poppin’, collaborative, and quietly competitive, all v Cube.
What Cubies Built (AKA: Real Problems, Real Solutions)
The best thing about the day was how practical everything was. People didn’t use the hackathon as an excuse to build fanciful things they’d abandon tomorrow; they went straight for the tasks that actually slow us down.
Here’s what came out of it.
The PM Meeting Scheduler
The “finally, someone sorted this” tool.
This team tackled a universal pain point: setting up recurring project meetings. Multiple dates, multiple people, multiple links, multiple updates, done manually every time.
So they built a tool that:
pulls project + team data from Forecast
generates meeting invites
updates them automatically when Forecast changes
It’s simple, clean, and genuinely useful. The kind of thing that quietly claws PMs back hours over the year.
The 3D Biomarker Viewer
From nightmare creature to functional prototype.
This team tackled a real prospect request: a 3D human model with interactive hotspots.
We gave them a pre-hackathon interview:
Early attempts produced the now-infamous “monster,” which absolutely deserved its own slide:
)
Once they stopped fighting with dodgy 3D generators and instead built the app first in Replit, everything fell into place.
By the end of the day, they had:
a working hotspot viewer
a rotating model
consistent visuals
a clear path to polish
It’s a great example of fast prototyping done right, and got our wheels spinning about Replit…
The Scheduling Assistant
The brave souls who entered “the Slack hellhole.”
This team wanted a bot that understood calendars, time zones, preferences, holidays… everything.
The ambition was spot on. The integrations? Less cooperative.
Slack + Calendar + GPT + n8n + Widget Builder = a spicy combination.
At one point they shared:
“Slack integrations belong in the 7th circle of Hell”
But by the end, they had a working suggestion engine.It occasionally offered 2:30am meetings, but hey, nobody or no bot, is perfect!
More importantly, they learned a ton, and it showed.
Github Gandalf
“You shall not pass” for bad test coverage.
Ed and Tyler built a GitHub Action that:
checks unit test coverage
compares it with QA scripts
blocks PRs that don’t meet thresholds
auto-writes missing tests using AI
It’s smart, elegant, and immediately useful.
Reporting Automations
Three reporting pains, three solid solutions.
This group took on:
utilisation
budget vs scheduled time
milestone reporting
Instead of endless sheet wrangling, they built upload tools that instantly generate the data and visuals they need. Hugely effective, no drama, no fuss.
Accessibility Checker
Upload a screen → get WCAG fixes + Jira subtasks.
A genuinely thoughtful build.
The tool:
analyses a screenshot
identifies accessibility issues
explains them in beginner and advanced modes
auto-writes Jira subtasks
It supports designers and developers in a meaningful way and strengthens the inclusivity of our work.
Figma vs App Comparison Crawler
A life-saver for QA + design.
Another excellent contribution: a crawler that:
captures app screenshots
compares them to Figma
highlights differences
exports outputs for designers
A brilliant time-saver for anyone going cross-eyed reviewing builds.
Demo Time: Cube Jazz Hands
By 4pm, we regrouped for demos.
Good builds, good vibes, and enough empty pizza boxes to suggest some serious Hackathoning had gone down.
We had:
polished demos
quick jokes
supportive roasting
plenty of “wait, you built that in a day?” moments
and one extremely cursed-but-loved 3D face
Without sounding overly effusive or cheesy, it was everything we hoped the day would be and more. A day spent rubbing elbows and problem-solving across disciplines with the most brilliant minds is a day extremely well spent.
And the winners are...
In true 3SC fashion, we couldn’t come to a unanimous decision, so we awarded two winners.
Judges’ Choice:
GitHub Gandalf by Ed and Tyler
A perfect example of AI elevating engineering quality.
Popular Vote:
The Accessibility Checker by Ginny, Georgia W, Juwon & Katy.
A thoughtful, creative build that won the whole room over.
👉 So… was it worth it?
Abso-frickin-lutely.
Taking one day out of delivery to build, learn, experiment, and fix our own pain points was absolutely invaluable. Teams came away:
sharper
more confident with new tools
with prototypes that will actually be used
and with stronger cross-team connections
This is what an AI-first agency looks like behind the scenes, no robots running the joint, but the everyday practice of carving out space to learn and improve together.
And yes… Replit made a lot of quiet magic happen. SPOILER: We have a blog out now covering what we thought about the no-code builder
Published on 4 December 2025, last updated on 4 December 2025