I come from test engineering. And watching Claude code always felt like a black box.

In my last post I wrote about creating a resume builder using

's 3M framework. Today i write about a realisation that i had which is in a way points in the direction of the WAT framework.

I come from test engineering background. Been doing V&V of AD/ADAS systems for about 10 years. A key aspect of being a Test Engineer is observability. Knowing what is happening under the hoods so that i can try to break it to ensure that it doesn't break in production.

During all my claude builds, Claude code always felt like a black box for me. I knew something was happening. But what, when, and in what order? No idea.

That changed today - and it happened because of one single line of code.

I was building an E2E test for a Scrum Master agent I've been working on for my OS. Mid-build, I had this nagging thought: I want to see the stages. I want actual logs. I want to know what phase Claude is in, what it's writing, and when.

So I added one line. Just one: tail -f /tmp/avya-os-e2e.log

And it spun up a live log file - a real-time window into exactly what was happening during test execution. Every stage. Every move. Right there in front of me.

For someone coming from a non-software engineering background, this was a genuine perspective shift. Claude stopped being a black box. I could see the build. I could follow the thought process. I could monitor progress like I was watching a dashboard.

Yes, you can always go read the code. But that's not the point. The point is observabiltiy without friction.

If you're building with Claude and you've ever felt like you're flying blind - this changes that.

The WAT framework gives structure and differentiates between the deterministic and probabilistic parts of AI. Being able to see what my agent is doing during the E2E test phase, gave me a way of precisely observing the deterministic part.

For the builders in the community - Would love to know if anyone else has been doing something similar. How are you keeping tabs on what your AI agents are doing behind the scenes? Let's swap notes in the comments!

0 comments

I come from test engineering. And watching Claude code always felt like a black box.