Testing and verification is where we all should and will spend most of our time; things that we dreaded are what we are going to do now :)
The “center of gravity” is moving from implementation to verification. So, now we need to decide how we handle this, both at the individual and org level.
Let’s start with the classic and then talk about an interesting one.
Some orgs are shipping at inference speed, trusting the model is good enough, while others have seen models produce wildly different outputs for the same task. Both things are true at once, and that is exactly why verification deserves more attention than it is getting.
The fix is not to slow generation down. We should instead speed verification up and automate as much of it as possible.
Wherever a check can be codified, do it. Linters, type checks, regression suites, test coverage, and CLI based testing can all catch classes of mistakes before a human even opens the diff.
If the software is meant to be used through a browser, test it the way people actually use it - clicking through real flows, checking states, failures, behavior, and the little UX details that unit tests will never catch. The goal should be end to end correctness, not just code correctness.
And then there is the part that is much harder to automate: empathy.
Use the product the way a real person will use it. Think about what they will misunderstand, where they will get stuck, what happens when they make the wrong choice, and whether the experience makes sense without knowing how the system was built.
Let automation catch the mechanical mistakes so that we humans can spend our attention on what actually needs judgment: intent, trade-offs, edge cases, and whether the change is the right one to make at all.
The bottleneck moved to verification, and honestly, that is a good bottleneck to have. It is the one that keeps a human accountable for what actually ships.