Building software with AI is not as easy as social media says. I learned that on day nine of my 30 day AI experiment, and it changed my business plan.

Today is consecutive day 2,153 of my daily run streak. I have run 16,534.66km so far, which leaves 23,540.34km to complete a lap of the world at 40,075km. The mission behind it is to raise £1M for children's causes. I run barefoot-style every day, and I film it as I go.

The experiment is simple to describe. I am taking one business concept through build, launch, delivery and, hopefully, profit within 30 days. I am doing it with AI, and I am doing it in public, so the result is honest whichever way it goes.

What happened on day nine

Most of today went on fixing features I had already built. That is important work. It is also the work that feels like standing still, because nothing new appears at the end of the day.

I have come to think this is normal in software development. AI lets me move far faster than traditional developers did, but the pattern is the same. You build something. You build something else on top of it. The second thing breaks the first. You go back, check for errors and repair what you had.

My honest read is that I am better than average with AI. I am not at the forefront. If I find a comprehensive build this hard, someone with less AI experience will find it close to impossible. That was the realisation of the day.

Why this matters if you are thinking of building a product

There is a lot of content online saying you can build a business in seven days with AI and make money from it. I do not think that is true for most ideas. It could hold if your idea is very simple, nobody else has thought of it and you are strong at marketing.

If you are solving a need you found in your own business, the software is probably complex. It needs a database, a back end, a front end and a clear structure. To build it you need to be logical and plan in a set order. Without that, you trip yourself up. You spend large amounts of time fixing bugs, and the output is often weaker than it should be.

So my lesson for anyone vibe coding a real product is this. AI makes building faster. It does not remove the need for structured thinking. Planning matters more, not less.

How it changed my strategy

My first vision was a set of small operating systems for businesses. Each would be packaged, mostly run by AI and sold for a one-off fee. I then moved to software as a service, because a subscription made more sense. Today I went back to a more consultative, higher-ticket model.

The reason is that what I have built is too complex to hand over with a set of training videos. I will still provide the videos. But implementation decides whether a client uses the system or abandons it. If onboarding fails, the business stops using it.

The model I am moving to has two parts:

- An implementation fee up front, so the client is set up properly. - A support subscription, where I act as their digital operations director. I add new KPI reports, fix workflow problems and automate new processes as they come up.

I still want AI to carry 90% of the workload and people to carry 10%. That has not changed. What has changed is how the product reaches the client.

There is a benefit to owning what I have built. It replaces many separate software subscriptions with one business operating system. It covers real-time financial reporting, team management, project management, planning and operations. I am also building a marketing element I never planned in the earlier version. Client feedback goes into the product, and I can switch features on or off per client so nobody is overwhelmed.

The timeline is also more honest now. I will keep running the 30 day experiment to see what it produces. But for business planning, I have set the target as live by the end of the second quarter, which gives me roughly 70 days. Thirty days now looks like a stretch.

The link to the streak

This is the same lesson the run teaches me every day. The work that feels like standing still is usually the work that holds everything up. Fixing what you built is the same as the run on a flat, unremarkable day. Nobody films it for the highlights, and it still counts.

My streak vision runs for 16.5 years. Over that distance, intensity matters far less than showing up in the same way each day. Software is the same. A realistic timeline and steady repair work will get further than a seven-day promise.

Every episode is part of a fundraiser for children's causes. The more people the vlog reaches, the more we can raise. That is the reason I publish every day, including the days where the update is that I fixed things.

If you are building something with AI, I would like to hear what you have found hard. Leave a comment or send me a message. If you believe in the mission, please subscribe, like and share the episode. Each view helps the fundraiser reach more people.

Tomorrow I run again, and I carry on with day ten.

Watch the full episode: https://youtu.be/FZ4kmAtk3dw