© 2009-2026 Rock Solid Knowledge
Thursday, 08 January 2026
Emma Garland
14 minute read
On Wednesday 22 October, the Rock Solid Knowledge Umbraco and web development team spent the day in the office for a lightning, AI-first hackathon, also known as a vibe coding session. The only rule was that everything had to be created using AI.
The idea for the day was in the works for a while, but after a surprise Halloween hackathon at our Town Hall event the day before, we learned that for AI developer, small groups worked best! We decided to keep that format and focus on experimentation and enjoyment, rather than competition.
We had clear goals for the day:
After a morning of scrums, catch-ups and co-working, we shared the list of potential AI app ideas on the big screen. Everyone talked through their favourites and added their names to the ones they wanted to explore. Knowing how the day before had gone, a team of people around one computer just wasn't going to cut it. By late morning, three pairs had formed, each ready to tackle a different idea using a mix of Open AI's Codex, Claude Code, GitHub Copilot, and Lovable AI tools.
We had pizza delivered for lunch, then kicked off two hours of focused "AI-first" building. What followed was a mix of teamwork, problem-solving and plenty of laughter (and sometimes frustration!) as each pair took their chosen idea from prompt to prototype.
Ross and Arianne wanted to create a calendar tailored for foraging enthusiasts. The idea was to display what plants, mushrooms and seaweeds are in season each month, with the option to log finds and get reminders for future seasons.
They started by using Word and AI to generate a full technical specification. Lovable, the tool they chose to build with, took that spec and produced a surprisingly polished website, complete with styling, navigation and lists of items by season. It even included warnings about poisonous lookalikes, which they had specifically requested to keep things safe.
The challenges began once they ran out of credits and tried to move the project locally. Connecting GitHub and running the code in TypeScript proved trickier than expected. With limited experience in that stack, they had to call in help. Karl Tynan stepped in to help troubleshoot, earning the nickname “organic intelligence” for his timely rescue.
Although the pair hit roadblocks when switching to GitHub Copilot for further development, the experience taught them a valuable lesson about AI compatibility. Some tools simply don’t understand each other’s code and knowing when to stick with one approach can save a lot of time.
Despite the technical hiccups, their finished foraging calendar looked great and worked as a first prototype.
Each month showed detailed foraging information, such as displaying seasonal examples (like wild garlic for April).
Specific detail pages also highlighted any potential safety and foraging warnings for each species.
The team felt it was a great start for an app, and the next piece of work would be inputting valid data from foraging books and dictionaries.
Karl and Shelly built an AI-generated website for the 2026 Formula 1 season. They chose Claude Code as their main tool, partly inspired by its success in the Halloween hackathon the day before.
Starting from scratch, they first instructed Claude to create a spec. After that, the pair’s aim was to develop without opening a single file. They used the command line to instruct Claude to create a static HTML and CSS site listing teams, drivers, and races. Within minutes, they had a working site with decent styling and functional navigation.
However, it became apparent that while visually the site appeared functional, the tool had duplicated markup across components such as headers. To improve maintainability, the team asked Claude to convert the site into a .NET Razor Pages app, complete with services and models for managing data.
The tool handled the conversion impressively well, completing the entire transformation in less than half an hour. It even restructured the JavaScript for caching and converted the styling to Tailwind CSS.
Not everything was perfect, though. Some of the race data was inaccurate, and the AI struggled to find reliable sources for the upcoming season. When corrected, it began sourcing information from other websites to fill in the gaps - and even recognised that a new F1 team would be joining next year.
The team’s main takeaway was that AI can produce strong starting points very quickly, but success depends on well-structured prompts and clear validation at every step. They noted that Claude also prompted them to use newer versions of technologies like .NET 9, which was interesting to see, even if they chose to stay with .NET 8 for stability.
For the third and final project, Kyle and Emma built a theme park generator. They initially decided to use a company ChatGPT Enterprise license, which leveraged Codex and GitHub integration. The goal was to create a site that could design entire theme parks, complete with rides, statistics, and names.
After an ad-hoc focus group session with expert theme park user Karl Tynan, they began with a specification that described the pages and features they wanted. Running it through Codex, the first version came back looking clean but bland. It had named the park Meridian Peaks, which they both agreed sounded more like a retirement home than a thrill destination.
Siloing the functional feature each developer was working on was the solution to avoiding merge conflicts; much like human-first coding. The team also considered their stretch goal to theme the site according to user preference, such as a nineties pixelated computer game vibe.
After some technical issues with GitHub connection, the team split into two to try and hit their objectives. Emma stayed on Codex and Kyle used GitHub CoPilot. Whilst this sounded more efficient, the sheer volume of merge conflicts and diverging solutions cost far more time than it saved.
The system could even generate new rides on demand through a "Ride Forge" page. This was a ride generator that stored the last six creations in local memory, producing names and stats based on user-selected themes and budgets.
Through prompting and iteration, the team refined the look and feel, developed a rides page showing queue times, height requirements, and intensity levels, and created an interactive map. The “map” was visually abstract, but it displayed consistent data across the site.
As a final touch, the team asked AI music generator Suno to compose a theme song for their fictional park, Bedlam Rockworks, using the styles they’d developed for the theme park: “Crustbreaker, Bedlam Rockworks, Bristol, industrial, theme park, instrumental, loopable, short”. It played through the browser and you can hear that it was LOUD.
Our team's main takeaway was that coordination matters as much as creativity. The speed and impact at which merge conflicts became an issue when both of us were prompting at once shows that AI collaboration still requires clear human communication
Rock Solid Knowledge
Throughout the day, Orestis Orphanou provided on-the-ground support, helping with setup, troubleshooting and any technical issues that cropped up. His help kept things running smoothly when the AI tools went off script.
By the end of the afternoon, each team presented their creations to the group. The demos sparked discussions about what worked best, how different AI tools handled development tasks and what could be improved next time.
Several themes emerged from the experience:
We enjoyed the day. It was productive, creative, and light-hearted. Each project demonstrated how AI can accelerate the initial stages of development, while clearly highlighting the importance of human insight and problem-solving.
The Umbraco and web development team left with working prototypes, lessons learned, and we have many ideas for what to explore next time.
Last updated: Thursday, 08 January 2026
She/her
Head of Operations
Emma is Head of Operations and a software engineer with multiple Umbraco MVP at Rock Solid Knowledge. She has a keen interest in AI integrations.
We're proud to be a Certified B Corporation, meeting the highest standards of social and environmental impact.
