One of this week's useful AI moments started with a photograph of Kattis's watch. I needed an English version of the image showing Marvin's message for a keynote. The result was good enough for what I needed, even if the lettering was a little too sharp.
It prompted a question about old AI disappointments. We can get stuck with an outdated picture of what the tools can do, while being dazzled by the newest spectacular demonstration. Meanwhile, a small practical task that failed before may have become entirely doable.
And perhaps the change is in us. Our own ability to use the tools may have caught up with the thing we wanted to do. That is a good reason to revisit the digital parking lot where the abandoned experiments live.
I still carry too many handovers
Today I found myself explaining how much coordination I still do between my AI collaborators. I start plenty of work. Keeping track of what has stalled, what needs passing on and what is waiting for me is another job altogether.
I've brought Todoist back into the mix as common ground for the team. I still love ticking things off. But I want the agents handling more of the task management and handovers, rather than leaving that work with me.
The foundation is a shared file repository: context, current status, changes and work in progress. Markdown files may seem a fairly basic way to organise an AI team. They work pretty well. What I'm trying to improve is how the collaborators pick up waiting work, update that shared memory and give one another feedback.
This remains work in progress. I need to spend more time improving the system and less time coordinating everything myself.
Follow a thought, without forcing it
This week I asked for meaningful connections between the posts in Notes. The instruction mattered: don't force it.
There are now reading paths connecting 45 of the 67 notes, with short explanations beside the links. You can follow the Four Jobs from the original explanation into practical change, or explore how old archives became material to build with. Direct follow-ups and related reading are labelled differently.
The old words and dates remain. The connections are a separate editorial layer, credited AI + JJ. Twenty-two notes still stand on their own. A useful trail should give you a reason to take the next step.
AI + JJ. A dated editorial account of the work and thinking, with sources linked in the text.