Vibe coding frustration rarely gets an honest airing, so here is mine. Day 21 of my 30-day AI experiment was the worst day for productivity of the 21 so far. I want to explain what went wrong, what I am doing about it, and what I took from the day.
I am taking a business concept through to profit in 30 days, using AI for 90% of the work. I talk through each stage on my daily run vlog. Today is consecutive day 2,178 of the streak, and my total distance stands at 16,713.47km. This leaves 23,361.53km of the 40,075km lap of the world. I run barefoot style every day and vlog every run to raise £1M for children's causes. The more people who see the episodes, the further the mission travels.
Simple things needed fixing. I wrote my prompts with care. I ran my audits and verification checks afterwards. The fixes still did not hold. One bug fix led to another round of iteration, then another. Each loop took a little more motivation from me. Once the motivation went, my sense of flow went with it, and my output fell sharply. I finished far less than I expected to.
Sometimes AI gets a task wrong, and the correction takes more work than the task deserved. I know this. Knowing this did not stop the frustration.
Compared with the old way of writing software, the pace is still enormous. Chalk and cheese. Yet I notice how quickly I adjusted to this pace. A small delay now feels like failure, even though the same build once took far longer. So I check my own expectations before I blame the tools. The speed is real. My baseline for what feels acceptable has shifted.
Social media shows the polished version. People describe monetising AI as smooth and straightforward. I agree the opportunity is real. The execution is slower, more repetitive and more tiring than most posts suggest.
For 21 working days I have looked at terminals and code screens every day. Checking everything works takes steady concentration, and concentration wears down. A tired mind decides more slowly and tolerates bugs less well. A low day after three weeks of this is no surprise. This is the cost of the pace I set myself.
Claude Code has a built-in command, forward slash insights. The command reviews your whole history with the tool, the messages you sent and the days you used it, then reports what you do well, what slows you down, and how to get better output. I ran the command once earlier in the experiment and made significant improvements to my productivity. I have plateaued again, so I ran it a second time today.
The recommendations are more technical now. I read this as a sign I have become more proficient, since the easy gains are already taken. I am working through the suggestions. If you use Claude Code and have never run this command, start here.
This week's main focus is the migration. I have not started, because I am still finishing a handful of modules. I aim to complete those, along with the bug fixes, in the first half of the week. Doing so leaves roughly half a week for the migration.
I apply the rule of three to my estimates. Whatever I think a task will take, I multiply by three, and I land closer to accurate. If the migration overruns, the marketing campaigns planned for next week lose time. This is the first moment I feel slightly off track. I am still confident of being live and launched inside the 30 days. More marketing time gives a better chance of a first sale, so I will see how the week unfolds. I will report the result either way.
I plan the run streak over 16.5 years. Across a span like this, low days are certain. Some runs feel heavy. Some days produce little. The streak does not ask for a good day. The streak asks for the run to be done. Consistency over intensity means a poor day still counts, even when the output is small.
Day 21 follows the same pattern. I did not hit a crazily productive day, and I do not expect to hit one every day. I accept the lull, learn from the data, and start again tomorrow.
Every run adds to the 40,075km. Every episode gives the mission a chance to reach someone new, and the more people who see the vlog, the further I move toward £1M for children's causes. I do this every single day, including the days when the work frustrates me.
If you are building something with AI, tell me where the process slows you down. Watch the full episode, subscribe, and share it with another founder in the middle of their own build. I will reply to every comment.
Watch the full episode: https://youtu.be/L_YYawdHNrk





