Today was day 8 of my 30-day AI experiment, and it was productive and frustrating in equal measure. I am taking one business concept through build, launch, delivery and, hopefully, profit in 30 days, with AI doing around 90% of the workload. Today was also consecutive day 2152 of my daily barefoot-style running and vlogging streak, which gives me a useful comparison for what I learned.

I did not start a single new module today. I spent the day testing, tweaking and fixing errors. It looked like very little progress from the outside. By the end of the day I thought it was some of the most important work of the whole experiment.

The frustration first. I did not get an update from my other two team members, which slowed a few things down. I have learned not to build a day around what other people might send back. I made progress on what was in my control and moved on. The day was still productive, but not in the way I had planned.

Now the bugs. You see a lot of content online saying you can make money with AI super easily. After eight days, I do not believe that. Yes, AI is doing around 90% of the build work, and I am not going to downplay that. But the work still needs a strategic, logical mind behind it. Vibe coding makes it very easy to go off on a tangent, start something, leave it half finished, move on to the next idea and forget where you were. Nothing about it is hard to start. Finishing is where the effort sits.

Testing is where this became clear to me. The strength of any piece of software sits in how robust it is, how many bugs have been found and how many errors have been repaired. Once you have paying customers, nobody expects bugs. If they do appear, customers expect them to be fixed quickly. A product that works most of the time is a prototype. So a day spent only on error fixing is not a wasted day. It is the day that decides whether anything built so far is worth selling.

What I have built so far impresses me, and that is my honest opinion. It also has a long way to go. There are still bugs to find and errors to repair, and I would rather say so plainly than pretend the product is further along than it is.

The second thing I realised today is about infrastructure. The software I am using at the moment suits a single business shared with a couple of team members. A saleable SaaS product needs more behind it. We will have to migrate to different software and put more infrastructure into the back end. I estimate that at about a week of migration work. The honest version is that it would probably take around ten weeks done the traditional way. Because we are building with Claude Code, and because of the speed it works at, a week is realistic with clear direction. The key phrase there is clear direction. The speed comes from AI. The direction has to come from me.

That leads to the lesson I keep returning to. The quality of your prompting matters. So does the logical pathway you follow. You need a plan, and you need a very clear picture of what you are trying to build. Without them, AI will build you something, and it may be well made, but it may not be fit for the purpose you wanted. AI does exactly what you ask. If you ask for the wrong thing, you receive the wrong thing quickly.

This is where the running comes in. After 2152 consecutive days, I know that a streak does not survive on motivation or intensity. It survives on a structure that does not depend on how I feel. I have run and filmed in good weather and bad, tired and fresh, and on days when little else went to plan. The approach is the same one I am applying to this experiment. Keep a structure, follow the plan, do the unglamorous work and finish what you start. A run does not need a new idea each day. It needs me to turn up. Fixing bugs is the same. Nothing about it is exciting, but it is the work that holds everything else up.

I should also be honest about the mental cost. Vibe coding is draining. Spending hours directing AI, checking its output, finding what is broken and deciding what to do next takes real concentration. You need the right headspace for it. I notice the days I run well and clear my head are the days I direct better. That is not a coincidence, and it is one more reason the daily run protects everything else I do.

Where am I overall? I still believe the experiment is ahead of schedule, and I still believe we will hit the 30-day deadline. It is simply an intensive process, and I would be misleading you if I said otherwise. Every day that passes makes me more convinced the outcome is possible, and also more certain that it is not easy.

All of this sits alongside the bigger mission. As of today I have covered 16,527.61 km of the 40,075 km I need for a lap of the world. That leaves 23,547.39 km. I am running barefoot-style every day to raise £1M for children's causes, and the vlog is part of how I get this mission in front of more people. The more people who see it, the more support the fundraising can attract. That is why I record every single day, including the days when the main topic is bug fixing.

If you are building a business with AI, my advice from day 8 is short. Plan before you prompt. Test everything. Finish what you start. Expect the work to be harder than the headlines suggest.

If you want to follow the experiment or the mission, watch the daily vlog, subscribe and share it with someone who would find it useful.

Watch the full episode: https://youtu.be/OLhgfzi567o