I keep coming back to the same question when I work on software:
What do we actually know, and what are we just assuming?
A green pipeline helps. So does a passing test suite. But neither tells me whether we tested what we actually shipped, whether the important path still works after deployment, or what happens when one part of the system stops behaving the way we expected.
I use AI a lot in development. It's fast, useful and occasionally very sure about something that isn't true. So I care even more about how a result was verified.
I've started using evidence-driven engineering as a name for this way of thinking.
Not a methodology. Just a preference for evidence over "should work."
- Evidence-driven engineering · v0.1.1 · 2026-10-05