On 11 July 2026, in the workspace and members' club of The Ministry in Borough, London, Rare Founders and Mollie held the Rare Build Hackathon. The event brought together a hand-picked room, capped at 150 builders, designers and operators, to turn an idea into a business in ten hours. AI Central covered the event as media partner, and the winning teams earned a feature in our newsletter. This piece draws on the organisers' published event page, their recap of the winning teams and the written answers of the ParkPal team.

Photo: Rare Founders
A few weeks before the day, the hackathon changed its own title. "From Prompt to Product in 10 Hours" became "From Idea to Business in 10 Hours". It looks like a copy edit. I think it is the most useful sentence anyone wrote about AI this summer.
Here is the argument of this piece. AI has made the demo cheap. When anyone can ship a working prototype before lunch, the prototype stops being proof of anything. The only proof left is a stranger who says yes: a letter of intent, a pre-order, a payment. Rare Founders built a whole day around that idea. Most companies still judge their AI pilots by the demo.
Why this matters now
Look at the partner list. OpenAI's Codex was the coding partner. Every seat came with AI access, infrastructure credits, database templates and a payment integration from Mollie, the Amsterdam-based payments company that co-hosted the day. Food, drinks and Wi-Fi were covered. In other words, the organisers removed almost every excuse a team could have for not building something.
That changes what a hackathon is for. Ten years ago, the hard part of a weekend build was getting the code to run. In 2026, a two-person team with an AI coding agent can stand up a working web app in an afternoon. The scarce thing is no longer the build. It is the judgement about what to build, and the nerve to ask someone to pay for it.
The organisers said it in one line on the event page: "We're not asking you to build a demo. We're asking you to build something people will actually buy." Then they added the rule that gave the line teeth: "One rule: no gaming the system."

Photo: Rare Founders
Rare Founders is a London founder community led by Vasily Alekseenko, with Fraya Brinkman as Partnership and Events Manager. It runs open-mic nights, demo days and co-founder matching evenings across the city, and The Ministry in Borough is its usual home. This was its first full-day hackathon. Vasily was clear in his invitation that they were "not only looking for tech people", because "you need a mix of skills on the team to build something great."

Rare Founders events across London. Photo: Rare Founders
The rules of the room
The format was strict, and the strictness is the story. Nobody could arrive with a pre-formed team. The day opened with matchmaking, and teams of two to four were locked before building began. The ideal team had one technical person, one design person and one commercial person.
The brief was open. There were no themes and no categories, and each business was judged on its own. Teams were told to leave the building to talk to customers. The judges counted four kinds of evidence:
letters of intent from the target audience
waitlist sign-ups
pre-orders
paying customers
Forms filled in by friends did not count. With Mollie's payment integration on every table, "paying customers" was not a figure of speech. A team could take a card payment on the day.
The day ran from team matchmaking through a full day of building, with lunch and dinner on site, to pitches from the top 10 teams, three to five minutes each with questions, then the winners' announcement and networking to close.

The Ministry, Borough, venue partner. Photo: Rare Founders
Four judges, four ways to say no
The judging panel was built to cover each angle of a business, and each judge's weight changed with the type of business on stage:
Business viability: Caroline Frankum, former CEO of Kantar
Investability: David Gilgur, Partner at Blue Lake VC, and Ken Thomas, Principal at BackFuture VC
Technical execution: Fergus McKenzie-Wilson, CTO and co-founder of 60x
Notice the ratio. One judge out of four looked at the code. Three looked at whether anyone would pay, and whether an investor would back it. That is the right ratio for 2026, and it is the opposite of how most internal AI reviews work.
What won
What the teams chose to build says as much as the rules. The podium did not go to a moonshot. It went to people who picked a small, annoying, expensive problem that millions of Britons already have.
First place went to OpenPub, built by Shutian, Shu-Hong and Komatsu.
Second place went to Fifteen, from the team V Squared, Vivaan Verma and Vivek Patel. Fifteen automates the tedious process of submitting Delay Repay claims: tell it your train was more than 15 minutes late, and it handles the rest with browser automation.
The market is not small. The Office of Rail and Road reports that train operators closed 2.5 million delay compensation claims between April and mid-July 2025 alone, and approved 80.2% of them. That is a large, boring, form-filling workflow, and boring workflows are where AI agents earn their keep first.
Third place went to ParkPal, from Tahmid Al Sifat and Faizan Alvi. It is a tool that helps drivers take the hassle out of parking penalties, by tracking and managing parking tickets and payments. Again, the market is large. RAC analysis of government data found that private parking companies issued a record 14.4 million parking tickets in the year to March 2025.

Tahmid Al Sifat and Faizan Alvi pitch ParkPal on the night. Photo: courtesy of ParkPal
We later asked the ParkPal team, in writing, how the day went. Their answers are the best account we have of what a ten-hour AI build feels like from the inside. On the first half hour: "We decided the problem, not the solution." They sketched the flow on paper (upload, extract, analyse, draft), limited the scope to "informational support only", and refused every debate about agent patterns, model choice or UI polish.
On the AI itself, they were precise. An open-weight model pulled dates and contravention codes from messy parking notices, but it "hallucinated confidently whenever operators used non-standard wording", so they built guardrails and rewrote much of the draft letters. Their summary is one I would print on the wall of every AI team: "AI accelerated everything, but only once we learned where to stop trusting it."
And on what broke: the agents passed half-parsed data to each other, and the team lost "about two hours debugging what turned out to be one missing comma". Then a Google API quota throttled them just before the demo. "Most of the time lost wasn't coding," they wrote, "it was chasing tiny failures that only appear when everything is glued together at speed."
What it means: our take
We see the opposite of this hackathon in most companies we talk to. An internal AI pilot gets a nice demo and a round of applause in the leadership meeting. Then nothing happens. Nobody asked the people who would use it to commit to anything. The pilot was judged on the build, which AI has made cheap, and not on demand, which nothing has made cheap.
Rare Founders fixed that with one rule. If a group of strangers can be asked for a letter of intent in ten hours, your quarterly pilot can be asked for one in ninety days. Ask the sales team to sign up for the tool before you build it. Ask the finance team to pre-commit budget if it works. If nobody will sign, you have learned more than any demo could show you.
The second lesson is the team rule. One technical, one design, one commercial. When the build is easy, the scarce skill is knowing who to sell to and how to ask. That is good news for senior professionals. Someone with fifteen years of domain knowledge is now the most valuable person on an AI team, not the least. Look at the podium again: train refunds and parking appeals are problems you understand from lived experience, not from a model card.
The third lesson is the ParkPal one: know where to stop trusting the model. Their product worked because they drew a hard line ("informational support only") and checked the output against it. Every company deploying AI in a regulated or customer-facing workflow needs that same line, written down, before launch.
For founders and for media, the lesson is about the room itself. Rare Founders turned a Saturday into a pipeline: talent for its community, users for its own products, and stories for partners like us. The winners got a feature in AI Central, the judges saw early deal flow, and Mollie put its payments product in the hands of up to 150 builders. That is distribution, not sponsorship. Later that autumn, Rare Founders and AI Central co-hosted The Distribution Edge in the same building, on exactly that question.
Key takeaways
The demo is no longer the proof: AI tools make prototypes cheap, so judge work by commitment from users
Ask for one signal: a letter of intent, a waitlist sign-up or a paid order before you call a pilot a success
Friends do not count: internal enthusiasm is not demand, so test with the people who will pay or use it
Pick a boring, expensive problem: the podium went to train refunds and parking tickets, not moonshots
Draw the line before launch: decide where the model stops and a human or a guardrail takes over
Mix the team: pair every builder with a commercial and a design mind, because domain knowledge is now the edge
Thank you to Vasily, Fraya and the Rare Founders team, and to Mollie, for a day that asked the right question. Next time you review an AI project, ask it too: who said yes?
AI Central reaches 300,000+ professionals across LinkedIn and newsletter. Share this with whoever on your team still judges AI pilots by the demo.
AI Central covers AI, tech and startup events, from global stages like London Tech Week to local meetups and founder dinners. AI doesn't have to be the only topic on the agenda. If there's an AI angle worth sharing, we bring it back to our readers
Running an event, big or small? Write to [email protected] and tell us about it



