This week I worked on the Vision and Learning Showcase with my project partner, Andy. We used ChatGPT and Claude for two different parts of it.

I used ChatGPT to refine the idea itself. Coming out of our Week 6 prototype and our Week 7 peer feedback, we knew our concept was too general. Working through it with ChatGPT, we narrowed our target user from any startup in any country to early-stage social enterprises in Kenya. It also drew a distinction I had not been making clearly, between whether an entrepreneur is eligible for a financing mechanism and whether that mechanism is actually suitable for their venture. Someone can qualify for a loan that would still be bad for their business. That distinction ended up being central to how we framed the prototype and our updated value proposition.

I then used Claude to turn that material into the deliverables. It wrote the narration script, built the slide deck, and drafted the written documentation from the material we had already written.

What stood out this week is that a lot of our presentation ended up being about places where AI had been wrong earlier in the project. In Week 3 we used Claude to simulate entrepreneur personas. One of them, Brian, was described as working on his laptop and later described as not owning one. The writing was convincing but it contradicted itself. The personas also assumed that trust was mainly about fear of scams. When we interviewed real people in Week 5, we learned that trust comes through relationships and institutions people already know. We decided to include both of these in the video instead of leaving them out.

There were also things to correct in what Claude produced this week. In the draft script it wrote that we had compared outputs across different model families. We had never named a second model, so that was more specific than what we actually did, and it was changed to say different AI models. The slide mockups also had placeholder values on the recommendation cards, like a blank interest rate and a blank institution name. Those are fine in a mockup, but they are exactly the kind of detail that would look real if we left them in without checking.