Sansi, I'm a fresh grad, and my company's business has little to do with AI. My daily work is just using AI to patch small issues.
I want to learn Agent development, but there are no projects at work.
Even if I copy an open-source project to learn, in interviews I can't answer whether my Agent is actually useful.
That's pretty normal.
At work it's rare to get the perfect chance — you want to build something, and someone just hands you a matching project.
Most of the time, you have to make do with what's already in your hands.
So what can I do in my situation?
Does your project have monitoring? Error monitoring, performance monitoring, that kind of thing.
Yes, we do.
Then first try using AI to analyze your monitoring data.
What does that have to do with Agent? One prompt is enough.
Keep going, and the connection appears on its own.
For example, when errors pile up, don't you first need to classify them?
Right.
Then can AI do that classification automatically?
That still seems like AI analysis, not really an Agent.
Right, so keep going.
Next, once AI determines what kind of bug it is, can it go ahead and fix it directly?
And if an error was already analyzed yesterday and appears again today, having AI re-analyze it wastes tokens, doesn't it?
Ohh, I get it. If I keep thinking this way, adding capabilities to AI one by one, it naturally becomes an Agent.
But is this kind of project easy to talk about in interviews?
I think it's easier than building an Agent demo from scratch. It solves a real problem, and you'll have data.
What data?
For example, how many alerts you used to get per day, all reviewed manually — how many does AI filter out for you?
How long it used to take to fix an issue, versus how long it takes now?
Ohh, that's great.
Also, monitoring is something almost every company has.
So in interviews you're no longer "I built a demo to learn Agent," but "I hit this problem, and to solve it I built an Agent."
I think the latter clearly carries more weight.
Got it. Thanks, Sansi.