AgentsInTheCloud - EvaluatingAgentWork

Everything is optimized to make it easy to evaluate the agent's work.

Let's start some work.

I'm going to go to my math tutor project here, change the red parabola to be a animated flashy rainbow and present it to me.

And then I'm going to go to my PI coding agent here and say I want to change the settings screen of the coding agent to be all animated rainbows because I'm interested to see what that looks like.

Going to pause while these are busy.

And they're done.

So let's start with this rainbow parabola that we kicked off first.

And the thing I want to highlight here is that both the parabola, is presented, but that it started this lesson not, in this beginning state where there are no parabolas, but that it actually navigated the app to the part where the agent's work can actually be seen and evaluated.

for the other one in the PI coding agent is the same.

Look at that.

it first made the change of actually changing the settings menu and then it figured like, how am I going to make Lucas life easier for him to evaluate?

And it decided to start the PI coding agent, manually navigate it to the settings screen.

And then because it knows the environment that it's working in, the agent, it can control my terminal view here and attach it to that same session.

So I can immediately observe what this sort of, you know, animated rainbow, if that is kind of the effect that I was looking for.

We have a few more features related to evaluating work and one of them is the changes view, which is a, git view.

Like, you know, I can look at the changes that my agent just did like this by selecting the working tree.

And we can see it like that.

We can also edit inline here if we want to get our hands dirty ourselves.

but this looks pretty good.

And you can also actually here in the history of the project, you can select a different range and even if there's like merges and stuff, you can say like, hey, I want to see the difference between here and here, which is a feature that I always miss when it's not present in Git Tools.

All right, so, agents are able to present the work to us to me using the browser, using the terminal and we're able to look at the source code changes conveniently with changes menu.

that's how it works.