In this section I review one AI-powered application and demonstrate how it can be used to create new value. This week is a bit different, because the application is one I built myself. And my focus in this post is on the gap between how it was made two years ago and how it was finished last week.
The first version
Two years ago I was just starting to play around with using AI to develop small tools I saw a need for. One of the first experiments was a Google Chrome extension to help me comment on LinkedIn posts.
Commenting on LinkedIn is a way you can have a strong effect on your visibility. Posting yourself draws attention, but thoughtful replies on other people's work build a reputation over time. A good comment takes a few minutes of genuine thought, and doing that ten times daily is unrealistic alongside everything else.
So I built something to help. At the time the AI development tools were nowhere near where they are today. I used Claude and ChatGPT chats, and I had to copy and paste the code they suggested into text files by hand. I had to keep the scope small, because every change might break something you cannot easily diagnose when you need to test things yourself.
I did get it to work. Mostly for myself, and for a few friends I shared it with directly. It sat there in that state for two years, useful enough that I kept it installed and rough enough that I would not have put my name on it publicly.
What changed last week
I decided to polish it and get it published on the Chrome Web Store.
This time it was much easier, and not because the model wrote better code. The difference was the size of the request.
I asked Claude Fable 5 to clean the whole thing up, run a security and usability check over it, test it thoroughly, and suggest anything worth adding. Not one fix at a time. The desired outcome, in one brief.
It found and fixed several issues, including ones I would not have known to ask about. Later that day I submitted the extension to the store.
Essentially, I gave it the wider job: get this extension ready for public release. That framing is the same argument I make about what long-horizon models ask of leaders, except here I have a shipped artifact rather than a theory.
Why the small version sat for two years
The extension was not blocked on capability. It worked. It was blocked on the unglamorous distance between something that runs on my machine and something a stranger can install without risk. Permissions handled properly. Errors that fail gracefully. A store listing that passes review. Someone else's security assumptions rather than my own.
Two years ago I could not have crossed the gap between a personal tool and a publicly released one, because crossing it by hand-copying code would have taken weeks of evenings. Last week it was an afternoon, and I never opened the code myself.
I have written before about vibe coding and what it means for executives who want to build things. The usual framing is that AI gets you to a prototype fast, which I covered in working backwards to a prototype in a week.
The tool itself
It is live and free. You can install LinkedIn Responder from the Chrome Web Store.
A caveat I would rather say myself than have you discover: this helps you respond faster, and it will happily help you respond blandly. A comment that reads like it came from a machine is worse than no comment. Use it to get past the blank box, then make the result sound like you, and cut anything you would not have said out loud. The judgment stays yours. What you are buying back is the few minutes of getting started.
Your action step
You almost certainly have your own version of the two-year extension. Something that works on your machine, that a few colleagues asked for, that never crossed the last mile. A spreadsheet doing a job it should not be doing, a script only you can run, a prototype from a workshop that everyone liked.
Pick that one this week. Write a brief for the finished state rather than a list of fixes: what it should do, who should be able to use it, what would have to be true before you would hand it to someone outside your team. Then give the whole brief over to one of the new models at once and review what comes back, rather than steering each step.
Time it. My guess is the distance you have been avoiding is shorter than the one you priced two years ago, and the only way to find out is to run it once.
If you want help working out which internal tools are worth finishing and getting them into people's hands, that is part of what I do in AI strategy advisory engagements and in hands-on build sessions as an AI keynote speaker and workshop facilitator.