This was one of the first agents that I created, all the way back in the beginning of 2024. Agent.ai was just getting started, and it was a very approachable way for someone like myself to build an agent. It was (and still is I believe) free. Can’t beat that price.
The two biggest challenges I had were generating longer quizzes and designing the UX.
I found that anytime I tried to create a quiz with more than 20 questions, it would just break. The reason being is that the context window was running out of space. It needed to load the transcript, create questions and corresponding answers, and output the entire quiz in HTML. I’ve since gone back and tested it with some newer models. It’s currently using Gemini 3, and can handle more questions. The last run I did was 30 questions and it handled it no problem.
For the actual quiz design, I just did a lot of prompt iterating. I was going into v0 and trying to reverse engineer based on screenshots of apps that I was trying to emulate. The ‘please give me the prompt that outputs this’ kind of thing. With that I got a good base prompt that I just kept iterating with Gemini. It was helpful to have the same model the agent uses also helping me iterate on the prompt. It just took a lot of time to get a pretty simple but reliable design like this:
There are too many projects too count where I’ve let complexity get the best of me. The thing I love most about this agent is that it just works. You just say how many questions you want, paste the URL, and then let it go. It was even featured as one of Agent.ai’s top agents of the week :)
You can check it out here if you want to give it a spin!