AI-Assisted iOS App Deployment

Welcome back. In this lesson, we are jumping forward to the first weeks around the Tokens Dashboard 1.0 launch. Instead of requiring a live App Store account, analytics service, or support inbox, everyone is going to receive the same fictional launch feedback dataset as part of the student materials for this module.

It contains test flight feedback, bug reports, App Store reviews, support messages, and release observations. Each record has an ID, so we can trace an AI-generated conclusion back to its evidence.

I’m not going to walk through every record. The important thing is that the dataset mixes bugs, requests, misunderstandings, positive feedback, personal preferences, and incomplete information. Our job is to organize it without pretending the evidence proves more than it does.

In this lesson’s student resources, go ahead and look at the AI deployment resources folder and check out launch feedback dataset. I’ll go ahead and open it in VS Code.

Here we have some product context, a fictional timeline, as well as test flight feedback, internal bug reports, App Store reviews, support messages, analytics, and release notes, and analysis reminder.

Now, we’re going to switch over to ChatGPT to begin working with our dataset. In the same project for Tokens Dashboard, I’ve started a new chat, and we’re going to go ahead and reference our student materials and open the lesson three prompts file. I’m going to copy the first prompt. And notice how we’re asking it to create a launch feedback analysis markdown file, and it’s going to use our launch feedback dataset.

As before, I’m going to tell ChatGPT to work on the following prompt and to reference the dataset file from within our modules folder.

Note that the student materials also includes a completed launch feedback analysis file that we’ve asked it to create as part of this prompt. But you can also use your own result, or you can choose to use the supplied sample for the next step once this prompt returns. Let’s go ahead and see what ChatGPT can do for us.

Perfect. Let’s take a look at the launch feedback analysis file that it has generated for us. Now, here we can see that there are some dataset scope and limitations. We also have an executive summary. So for example, it mentions that there are two accessibility defects which are confirmed, that there’s a third confirmed logic defect that occurs for empty model data. Hmm, our users are also repeatedly misunderstanding the watch label, and there’s really no conclusion about broader adoption, engagement, crashes, performance, or business impact that’s supported by the supplied evidence. Okay.

We have also a section on findings group by classification. So here it says confirm bugs, and it lists out quite a few possible bugs for us to review, messaging confusion that our users are not too clear on, as well as some feature requests. We also have a positive signal section to give us some good news and not just focus on the bugs or the ambiguity or things that may be unclear for users, which can kind of bring us down a little bit.

All in all, I think this has been a really useful exercise. And remember, you can then review these elements, go through them one by one, challenge the AI whenever you know there shouldn’t be that bug, you know you fixed something, or something has been updated, have it maybe review your code, validate that things are still actually bugs or ambiguous elements that maybe your users are reporting. Again, AI is here to help us, but not to make decisions for us and for us to just sit back and do everything. If you trust that the AI is doing a good job, go ahead and let it. But if you know something is not what it’s reporting, then go ahead and fix it, challenge it, or review things on your own.

Now, what we want to do is ask it to perform a review, if you want, in another conversation. So we have this feedback analysis review file that we can reuse, and we also have the prompt that includes these review instructions. So I’ll go ahead and show you the file that I am referring to in our AI deployment resources. In this feedback analysis review file, located here, is a sample file that you can use as a guide for future prompts. But again, this lesson’s prompt, the second one, already contains these instructions. So I’m going to go ahead and copy it.

Then I’ll paste it in ChatGPT. I’ll first close this markdown file. And in this case, I don’t want to start a new chat. I’m going to ask it to be impartial and to not use the context from this conversation to review and provide results and conclusions, but rather to do an impartial review and process the data just via the prompt. If I switch back to VS Code, you may see that it is expecting the launch feedback dataset file and the launch feedback analysis file. So I will also tell the prompt to look for those in our modules folder.

Great. So now we have the results, and let’s take a look. It has some adversarial review findings, as well as some elements for us to triage. And remember, you don’t have to take everything at face value. You can reject elements, defer, mark items as completed, or things that you’re just not really going to care about, worry, or implement.

You also will likely not have the exact same output that I do, and the important thing is not to focus on the output, but rather that you are getting similar results and being able to use your launch feedback dataset and your actual analysis to provide you with actual, tangible, actionable elements or things that you can do to mitigate some of these issues or address feedback or requests and suggestions that users may be sending your way.

Now, with that, we have an evidence-based backed analysis and a human-reviewed action list. AI has helped us organize the feedback and challenge the first result, but it didn’t make a final roadmap or set of decisions for us.

Across this module, you have prepared launch content, created an App Store listing package, and turned early feedback or bugs into useful next actions. The big takeaway from this module and this lesson is pretty simple: Just use AI to move faster through messy or cumbersome material, but keep humans responsible for what gets published, what gets built next, what gets fixed, and what recommendations you actually want to address and implement, and which ones you just don’t really care about. Now, with that, I wish you the very best, and thanks for joining me. I will see you next time. Bye!