I finished AI Agent Fundamentals with Azure AI Foundry course on Coursera, and then a chatbot that couldn’t cancel my cable gave me evidence of the examples of AI agents the course taught me.
Actually, both things taught me the same lesson from opposite sides. One was a course about designing agents that actually help people. The other was an agent doing the exact opposite.
The course that made me switch to classic mode
I’d abandoned this course back in March. I got stuck on some Azure setup in the first few steps, spent half an hour fighting it, and gave up. This time I went back, followed the same steps, and it worked. Whatever had changed, I got through it.
The course had four modules. I enjoyed the ones where I had to think through a problem and design an agent from the perspective of the person seeking help. What kind of AI service, what considerations, what would actually help someone. The course asked me about my thinking and nudged me when I’d missed something. That part clicked.
The UI did not cooperate. The screens in the video didn’t match what I saw in Azure AI Foundry. I tried the built-in AI chat to find where the options had moved. It couldn’t tell me. Neither could I. Eventually I asked Gemini, and it pointed me to classic mode. There it was.
A course about giving AI agents a better experience, and the built-in agent couldn’t help me navigate its own interface. I thought that was funny.
The chatbot that wouldn’t let me leave
The same weekend, I wanted to cancel my cable with Xfinity. I went to the website, found my account, picked the cable TV package. The options there were all upgrades to more expensive plans. Nothing let me cancel the current one. The search found nothing. The contact link put an AI chatbot in front of me.
I asked it how to cancel. It gave me a runaround. It kept telling me I could manage things on their website, and every link landed back on the same page with no cancel option. I wanted the number to call a human. It said it could help me. It couldn’t.
Note: I needed more assistance. It never put me in touch with an (human) agent.
Gemini had a different answer. It gave me the number, the information to have on hand, and how to make the request. Then I called.
A person answered. I asked to cancel my cable. They asked why, then came back with an offer to switch to a different package with more channels. I let them finish and said I’d already told them why I was calling. They canceled it. Twenty minutes on the phone to do the thing the website and the chatbot refused to do.
What I’m learning
I can’t get mad about this and expect anything to change. So I try to pay attention instead. When I run into bad UX, in software or in an AI feature, I screenshot it, save it in my notes, and write down why it felt bad. If I have an idea in the moment for how I’d fix it, I capture that too. If I don’t have time, I leave it and come back later.
I’m starting to look at AI chat and AI-powered features the same way. I’m the user of something someone else built. I’m paying attention to what helps me and what doesn’t, so that when I design something for other people, I don’t repeat the same mistake.
The course taught me to think from the person’s perspective. The Xfinity chatbot showed me what it looks like when nobody does. I’d rather fail and learn fast than fail and learn nothing. Actually, I want to learn fast so I can improve fast. That’s the whole point.
Leave a Reply