Agent engineering is where software engineering was before unit tests existed

We're launching PandaProbe Cloud, but I keep coming back to this thought:

Software engineering spent decades building the infrastructure to trust code in production — unit tests, CI/CD, canary deployments, logging pipelines. Nobody questions that investment anymore.

Agent engineering has none of that yet. We're shipping systems that reason, plan, and act — and our main debugging tool is still reading logs and hoping.

I think we're at an inflection point where the teams that invest in understanding agent behavior early will have a massive advantage over those who don't — the same way teams that adopted testing culture early won in the software era.

Curious if others see it the same way — or if you think agents are fundamentally different enough that the analogy breaks down?