# Transcript

**Anand**: [00:03] To begin with, does anyone have any questions from what we covered yesterday or what you were going to be doing yesterday? Because today's session is mostly going to be just answering any questions you might have, plus maybe just one additional task beyond what we covered yesterday. Okay, let me share my screen and let's see where we are on this.

**Johan**: [00:49] Maybe one short question, that's all.

**Anand**: [00:52] Yes, Johan?

**Johan**: [00:53] Yes, you mentioned that we have to iterate and use AI to verify the products, but **how many times should we do that until [it's done]?**

**Anand**: [01:00] So there are two parts to the iteration. One is the automated test case, and that—at least once, but probably no more than twice. I'd suggest you do it once to get a feel for how it works, and then once you've finished your product features mostly, then **keep this as an automated test set so that later on when you change something, the test still works.** The evaluation as different personas, that I'll leave entirely to you. At least once, but you can do it again, change features, do it again, change features. **There's no end to user testing.** Short answer for both is at least once.

**Anand**: [02:08] Let's go back. All right, we have 13—okay, well, 12 people submitting 13 entries. That's perfectly fine; each of you can submit multiple responses. And the agent research is available for many of you. I would encourage everyone to complete it, but let's take a look at any one of these to see what the agent research revealed. Maybe a couple of these.

**Anand**: [02:53] Okay, wait, that's again on ChatGPT. Let's take something that's not... okay, that's a Claude artifact. Any other platforms? That's a ChatGPT site as well. Okay, that's a GitHub page. So let's take them in that order. And this used a freelance subtitler, a municipal records clerk, and a mechanical engineering student. Interesting to look at how the walking application might be used by creating a use case and identifying friction. Okay, "If the phone is locked but the page is visible, that can be demanding on battery life." Yes. "Hesitations because of corner travel." Yes. "Inter-application coordination can be an important friction." Yes, that's a good set of items. I'm curious if this—okay, that's the link to the application itself.

**Anand**: [04:21] Route Draw takes a batch delivery coordinator and finds that, okay, the earlier route gets deleted. A retired Google Maps user finds that clicking gives no feedback. Good point. And a gig bike courier finds that the routing calls go from the browser to OpenStreetMap with no backend. Ah, okay, but it might get limited upon high use. That's a fair point. And then there are details. Great.

**Anand**: [05:11] And this is from a university student, as well as a professional studying English and someone restarting bookkeeping, exploring their study habits. Great. Now this is exactly the kind of research that we want people to do. So that is useful; do complete it. Did anyone have any thoughts or have any questions while doing the agent research? Or what did you learn from this? Just curious if you have any inputs.

**Anand**: [06:09] Do reflect on it because that's one of the things that I'm hoping you will be in a position to share in a supplementary video tomorrow. I mentioned that for tomorrow we'll have a series of—well, you'll be creating a video walking through your application. That's one of the outcomes of this session. But the other will also be an exploration of how we used these applications and agents, what we learned from that, what it did well, and what it didn't do well. No right or wrong answer; it's just walking through our experience and our takeaways from that.

**Anand**: [06:59] Then there is feedback for many of these, not all. Let's see. So, Johan, you have—okay, one feedback from Noah, not yet from Kosei; you might want to ping. KK hasn't filled in the—oh, I thought KK—oh no, KK didn't have an app yesterday either. We might want to nudge if he wants to complete that, but let's see. Hinata has at least one feedback as well. So does Nanami on both the applications. Dora does not—okay, Dora does have feedback. Blue, that's at least one. So does Noah. Okay, and Mio. Fair. So everybody has in some shape or form at least one feedback that they can incorporate.

**Anand**: [08:08] Let me just take a look randomly at one of these from Streamable. Okay, is it my audio that needs fixing? No, there is no audio. Okay. So this is a walkthrough just showing what they did, where they clicked, without a voiceover. Probably should have made that more explicit; it always helps when people hear what you're thinking. But fair enough, at least there is a walkthrough of how we did it. Let's take another one.

**Video Audio**: [09:17] ユニクロを選択します。 (I'll select Uniqlo.) [09:27] ここまで作業してみての感想... (Impressions after working up to this point...)

**Anand**: [09:32] Fair enough, and five minutes is a reasonably good duration to get a sense of the inputs. Got it. Perfect. So, let me stop sharing for a bit. **What we've done so far is taken an idea, which can be agent-generated, implemented a prototype, which again was agent-assisted, gotten feedback from agents and people, and now are using agents' help to create a revised version of it.**

**Anand**: [10:21] **The last stage is to then—for lack of a better word, I'm going to say—market it.** That is to tell the world: "Here's an application, here's how you should use it," and start trying to see if people are in fact using the application and learning from the process. So what I will be asking you to do is—and I'll be sending an email as well to everyone as a next step—launch this as a product. And when I say product, it doesn't have to be paid; a free product is perfectly fine. **See if you can find out who is using it and how, and learn and improve from the process.**

**Anand**: [11:13] Right now, we asked people to record a video; we asked an agent to try it out. Now we're going to ask unknown people to try it out. And therefore, **your application needs to have some mechanism to figure out who's coming in, or even if not who's coming in, what are they doing?** What's working well for them? What's not working well for them? How do you do that? Ask the agent. There are several tools out there. The "how to do it" is not the important question anymore; it's the "what to do" that starts becoming important.

**Anand**: [11:53] So that's one deliverable for tomorrow. And how will we be testing it? Well, I'll visit the application, you'll all be visiting the application, and as we use the apps, you should get some feedback that the application has been used in this way, and you should be able to learn from that and improve as the next iteration. That's one deliverable.

**Anand**: [12:22] **The second thing that you certainly should be doing is creating a two-minute video explaining the application.** I mentioned this yesterday as well, just a walkthrough of: "Here's what my application does." This may be helpful for you when you're publishing this. And when I say "publish" the application for others to use, I mean maybe put it on some social media site, your blog, send it as an email to people—however you want to socialize it. And I'm not setting any targets on how many people should read it or click on it, nothing. Just share it with your parents; they might not open the email, that's okay, just share it. That's perfectly fine.

**Anand**: [13:05] But we will be using it tomorrow—and we will be, I will be—watching all the videos offline tomorrow, and I will be using the application. When I do, you should know that somebody's used it in some way and use that to revise and improve the output. **Lastly, I'd also like you to share a short, two-max-three-minute video on what you learned from the usage of this.** Meaning how you went about using the agents, what it did well, what it didn't do well, and what you are taking away from that.

**Anand**: [13:50] I'd like you to support this with the shared transcript of one of the sessions where you were interacting with the agent for a long time. What I'm going to do is explain how you might be able to share the logs from these sessions. One option is, if you were having a discussion directly on ChatGPT or Claude, you can just click on the share button. Let me share my screen first and let's go to ChatGPT and Claude.

**Anand**: [14:38] So if you had a session where you were asking it to write—and do the survey, for instance, or were making changes—just click on "Share" and make sure that you select "Anyone with the link," save it. That will give us a link when you click on "Copy public link." And I will add a column to this sheet which is called "Session Link," and you can add a link to that session out here. This will give a sense of how you've been prompting it. The process is very similar for ChatGPT; pick your chat and then on the top right, you have a share button and you go through the same process.

**Anand**: [15:40] For Codex and Claude, it's a little trickier. Actually, I'm not going to explain this because you can just tell the agent to export the session and it'll export the session. Frankly, that's easy enough to do. So I will just leave that as an exercise for you to try. And if anyone's not able to do it by tomorrow, we are in any case meeting face-to-face and I will let you know how you can export this.

**Anand**: [16:09] To recap, what we're doing for tomorrow is:
0. (Of course, I should say) Update your applications based on the feedback to get it to the point where you think it's better.
1. Create a video of you explaining your application (two minutes).
2. Create—what was the first one that I said before that? Explaining the application, video of your learnings... I completely blanked out. Does anyone remember what—I said two videos and one other thing.

**Johan**: [17:02] Okay, so the first step was—the two videos of explaining and then afterwards was the... integrating how we know if somebody has accessed our [app?].

**Anand**: [17:39] Thank you! Yes, you're absolutely right. **Integrate analytics into the application, making sure that when somebody uses it, you know how they've been using it and can improve.** And the last of course is a video of your learnings from what you asked the agents to do, what it did well, what it didn't do well, and therefore how you should be improving it, along with a link on the same Google sheet which has one of your longer chat sessions.

**Anand**: [18:17] With that, we would have perhaps in tomorrow's class, if not at least by end of day tomorrow, wrapped an entire AI product development life cycle. And you should absolutely publish this on your portfolio. Any questions or clarifications?

**Johan**: [18:49] So for the video that publishes our findings, like what went well and what didn't go well and how we iterated the AI, it should be two to three minutes long, right?

**Anand**: [19:06] Yeah, keep it to that.

**Johan**: [19:08] Okay, did I cover everything to be included in the video?

**Anand**: [19:12] What you did, what it did well, what it didn't do well, and what you learned from that. That mainly. So: "Here's what I think it does okay, here's what I'm going to do going forward, here are my takeaways," however.

**Johan**: [19:34] Okay. And all of them, besides the transcript of the chat, are sent through the email?

**Anand**: [19:42] Ah, no, good point. Let's just add it to the same sheet. That is actually an excellent idea. Let me share my screen. So I'll put all three deliverables. I'll say "Product Video Link," and then "Learnings Video Link," and "Chat Log Link." And order-wise maybe I'll put this one here and we may not need this column. Yeah, just adding it here will suffice. Thanks for flagging that, Johan.

**Johan**: [20:36] Okay.

**Anand**: [20:37] Anything else from anyone? If there is, please just mail me, same ID. And otherwise, we will in any case be meeting physically, same room, tomorrow same time. Thanks, everyone. Have a good day.