AI debugging: from hypothesis to verified fix
The latest models make investigation and repair more capable. The real speed gain comes from shortening the evidence loop without weakening the definition of done.
Field notes
Concise observations on architecture, reliability, delivery, and the organisations around the software.
The latest models make investigation and repair more capable. The real speed gain comes from shortening the evidence loop without weakening the definition of done.
Treat architectural significance as a gradient: spend design effort where reversal, evidence, and accumulated coupling make change costly.
The useful output of architecture work is not a diagram. It is a set of decisions a team can act on, challenge, test, and revisit.
A struggling programme usually has capable people trapped inside a broken decision and delivery system. Start there.
Reliable systems come from architecture, operations, incentives, product decisions, and incident learning working as one system.