I think I should let go of the mental model behind ADRs.
ADRs carry a mental model that resists overturning past decisions.
I've written an ADR every time I made a decision, and while that gave the model rich context, it also held back its freedom to think.
Current models already lean toward safety and avoid taking risks.
Add the ADR mental model on top of that, and the tendency only gets stronger.
So I'm dropping the ADR format and rewriting all of them as plain decision records, with an instruction telling the model they don't bind future decisions and are simply a record of what was decided before.
It may be just a matter of labels.
But LLMs are very good at identifying what something is, so I think this does more than relabel.
Future observation should bear that out.
I'm telling you folks, ADRs and primitives are a huge leverage point for getting your agents to build the right thing. I don't have to worry about an agent going against an important architectural decision because they actually do read those and make the right choice.