This morning I spent 2 hours measuring how much work my own system had done — counting it, sorting it into categories, and then building a small tool to count it better.
My business earned nothing from a minute of that, and I only noticed when I stopped and asked what it was for.
A year ago that could not have happened, because a year ago I could not have built the tool that wasted the morning. The constraint was hard and it was honest: there were things I wanted and could not make, and the wanting had nowhere to go. It sat there.
The work got faster. Deciding what was worth doing did not.
I opened it in February 2025 not knowing Python and not knowing what Git was. Three months of that before it cost me a penny, and the first bill, when it came, was £18. I was not trying to build a system. I was trying to stop renting tools that broke and then charged me for the privilege.
What replaced it is stranger and nobody selling you an AI subscription is going to mention it.
When you can build almost anything in an afternoon, the deciding is the whole job. Not the building — the choosing. And choosing is the part that has never got any easier, because it was never a capability problem in the first place. It was always a taste problem, and taste does not have a subscription tier.
So the failure changes shape. It stops being I could not do that and becomes I did that beautifully and it was the wrong thing, which is a much harder failure to see, because it comes with a receipt. There is something to show for it. It looks like a productive morning.
I have watched this in my own work all year and the pattern is boringly consistent: the tool is quickest at the jobs that are easiest to specify, and the jobs that are easiest to specify are usually about my own setup rather than about a customer. Tidy the system, improve the process, and build the thing that measures the other thing. Every one of those is a real improvement, none of them puts a pound in the till, and the speed makes them so cheap to reach for that a month can go by before you look up.
It cost me a client. Not to a competitor — to impatience, one autumn, while I was heads-down building the thing that was going to serve them better. That does not appear anywhere in the accounts.
If somebody is selling you AI for your business right now, the pitch is about speed. It will do in an hour what took a week. That claim is broadly true and it is not the question. The question is what you are pointing it at, because it will build the wrong thing at exactly the same speed it builds the right one, and it will never once tell you which it is doing.
That is not a reason to avoid it. I would not go back. It is a reason to spend the first hour on the deciding rather than the doing, which is precisely the hour that feels like it is not real work.
The check I would run before spending money on any of this is not about the tool at all. Take your last month and ask what actually moved for a customer. Not what you built, or tidied, or automated — what a person outside your business would have noticed. If the honest answer is thin, more speed is going to make it thinner, faster.
It is the same instinct as deciding whether to fix your website or rebuild it, where the answer is sometimes neither yet, and the third option is the one nobody offers because nobody makes money from it. It is why improving what you already have usually beats starting again — the hard part is done, and doing it again feels more like progress than finishing it does. And it rhymes with the dashboard problem, except the dashboard at least belongs to somebody else — this one you build for yourself, out of your own good intentions, on a Thursday.
There is one more thing I have left out, and it cuts against everything above. The most useful piece of kit I own got built the exact way I have just warned you about — badly, 8 times, over 5 months, with nobody asking for it. I have written that one up properly on the other site. What I still cannot square is that the version which finally worked was a question I could not have asked at the start.
