WorkOS

Everything I needed to keep on top of was scattered across too many places: my own to-do list, the things I was waiting on other people for, notes from conversations, and the running threads across the different areas of my life. Each of those sat in a different app and none of them talked to each other.

What I wanted was one system, entirely local to me, that would take whatever I captured in whatever form I happened to capture it, organise it without me having to tag anything, and give me a view of how I actually work.

This was my first go at building around a local LLM, and the design idea was a blank sheet: drop anything in, in any format, and let the model do the sorting. It held a picture of how my life is structured, meaning the areas I'm responsible for, the people that work around me and which of those areas they sit in, so when something arrived with no tags or metadata attached it could read it and place it correctly.

The front end was a dashboard, with everything categorised down the left by person and by topic. Before a conversation with someone I could click their name and see every relevant note and every outstanding thing attached to them, and if I was thinking about a particular topic I could do the same thing by subject. From there I could take anything that needed doing, hit delegate, and it became a task sitting against that person, which meant I could go into anything with the relevant material and the outstanding work already in front of me.

The local model handled the categorisation well and the interface did what I wanted, so the build itself went smoothly, which is part of why what happened next caught me out.

I used it for a while and then stopped, not because anything broke but because I didn't want to maintain it. It had quietly become another system to stay on top of and another place I had to remember to go, so a thing I'd built to remove effort had added some.

That's where a bigger realisation landed. When the possibility becomes that you can build anything, the temptation is to build everything, and I'd built this because I could rather than because I'd genuinely established that I needed it. Because I built it, I was also carrying the mental load of holding its logic in my head as it went along, which is a cost you don't notice at the design stage.

The question to put to anything before it gets built is whether I need it, not whether I can build it, not whether it would be useful, not whether it would be interesting to make. I skipped that question here and paid for it with maintenance and a system that ended up collecting dust.

Every organisation going through AI adoption will run into this, because the pattern is excitement, then possibility, then building, and the building is now cheap enough that nothing naturally slows it down.

Without strict rules about what gets built and why, you end up with a graveyard of things that all need maintaining and deliver very little real value between them. So the governance question isn't only what we should build, it's whether we need this at all, asked brutally and asked before anyone writes a line of code.