The software with an addressable market of eight people

For six years, a group of my closest friends and family have run an Arsenal prediction competition. Predict the result, earn points, argue about the rules, and hand over a real trophy every May. It is exactly as serious—and as ridiculous—as it sounds.

The entire competition lived in chat. Predictions were posted manually. Scores were calculated manually. Reminders depended on somebody remembering to send them. Every season added more history, more rules and more reasons why a generic prediction app would not quite fit.

I had thought many times that I should build an app for it. But that sentence used to be the end of the idea. I am a product manager, not a software engineer. Commissioning bespoke software for a private football competition would have cost far more than the problem could ever justify.

RedLine did not suddenly discover a large market. The addressable market was still the same eight-ish people who had been predicting Arsenal scores for years. What changed was the cost of turning a precise intention into functioning software.

Generative AI made it rational to build something that would previously have been economically absurd. I could provide the product direction, rules, requirements and judgement while using AI to assist with the technical execution. The result did not need venture funding, a business case or thousands of users. It only needed to be useful to the people it was made for.

There must be millions of tiny problems like this: products nobody would ever fund, commission or build professionally. Now, maybe they do not need to.

The earliest build made the main flow possible: choose an Arsenal fixture, submit a score and calculate points. That was the demo. The product emerged from everything the demo ignored.

Our competition has bespoke scoring, power-ups with downside as well as upside, private leagues and years of history. Predictions need to remain hidden until kickoff. They must lock at the exact moment a match begins—even if somebody has an old page open. Fixtures and final results should arrive automatically. People who forget need reminders. One account owner may need to manage a child’s prediction without weakening privacy for everyone else.

None of those details looks dramatic in a screenshot. Together, they determine whether the people using the product trust it.

• Authentication and private league access
• Custom scoring and power-up logic
• Automated fixtures, results and reminders
• Prediction secrecy and hard kickoff boundaries
• Historical competitions and managed-player support

Saying that I built RedLine with AI does not mean pretending I wrote every line of code. It means being precise about where the value came from.

I knew the problem, the people, the six years of accumulated rules and the moments where the competition could become unfair. I translated that context into requirements, challenged weak outputs, made trade-offs, tested the real experience and kept iterating when the first answer was technically correct but practically wrong.

AI dramatically lowered the technical barrier. It did not decide what deserved to exist, what ‘fair’ meant, which edge cases mattered or whether the experience was good enough to replace the chat. Those remained product decisions.

Once friends and family used the product, the comfortable version of the story ended. Testing exposed spacing problems on mobile, unclear states, mismatched historical tables, reminder questions and cached pages that could threaten the kickoff rule. The main flow worked; reality arrived through the edges.

That process changed how I think about AI-assisted building. The speed of the first build is impressive, but speed also gets you to the difficult questions earlier. The advantage is not that AI eliminates product work. It gives you a real product on which to do product work.

Traditional software economics pushed builders toward large markets. If a product required a team and a major budget, it needed enough customers to support both. That logic filtered out countless specific, useful ideas before anyone could test them.

AI does not make every tiny idea worth building. It does make more of them testable. A family competition, a personal decision system, a niche workflow inside a small business—these can now become functioning products without first pretending they are the next enormous platform.

RedLine may never need more than its original group. That is not a failure of ambition. It is proof that software can now be proportionate to the problem it solves.