Ep. 63: What Should Your AI Have Access To?

[00:00:05] Welcome to Prompting Curiosity, a podcast for the AI curious. No coding background required. I'm your host, Dr. Shantae Cofield, also known as the Maestro, and I created this show to explore what these AI tools actually are. Really, though, are the files in the computer, how to use them, and what they might mean for how we think, work, create, and move through life. Whether you're skeptical, intrigued, or already experimenting, you're in the right place. All that I ask is that you stay curious. All right, let's get into it. [00:00:38] Hello, hello, hello, my curious people, and welcome to episode 63 of Prompting Curiosity. And also welcome, um, to October. I am your grateful host, the Maestro, and today we are talking about what your. That's the emphasis there, what your AI should have access to. So, fair warning, this episode's not, you know, it's not about me telling you what to do. I, uh, I really want to use this episode to highlight how I believe AI sentiment has changed this incredibly over the past four years, and also offer up some thinking points, slash questions that you can ask yourself to determine what you want to do with your AI. Right. Of, uh, note, I'm going to just emphasize that again, I am specifically talking about your personal AI and your personal software tools and information. I'm not discussing what I think these AI companies should or should not have access to. [00:01:30] I don't really care to talk about that. Um, and double note, ah, you know, as I'm going through this, this episode, if you're like, I don't even know how to connect AI to my other software, as per, always simply ask the robots. Okay, so into the main topic here. When, which is what should AI have access to? What should your AI have access to? So if you back it up here, when chat, when chat, GPT hit the scene way back in November of 2022, which is wild. It's almost been four years. It's crazy. [00:02:05] Uh, there were really no concerns that I can recall. I can't speak for everyone. I'm gonna speak for myself, but I'm, uh. There are no concerns that I can recall as related to access, namely because, like, it didn't have access to things. So there may have been some what ifs, like, what if AI gets access to this? You know, but generative AI was not actually connected to any software or inside of the tools that we use on a daily basis. It was out there just writing poems in, you know, different voices. Right. [00:02:35] It wasn't until a year later. So that's roughly late 2023, when the ability to even connect AI to your software became a thing. And this wasn't even a thing, like relief to the public. This was to, uh, to those that used Bard. [00:02:51] You ever heard of Bard? Yeah, me neither. Um, so Google's original chatbot was called Bard. And as I was doing research for this episode, I was like, oh, my God. Yeah, I do remember rumblings about this. [00:03:03] Um, but I wasn't following the AI stuff as nearly as closely as I was as I am now. And, um, also I was using ChatGPT. I was not all up in clawed, um, and definitely not using Google. So, uh, 2023, Google made it so that Bard could pull information from Gmail, docs, Drive Maps, YouTube, uh, and Google flights and hotels. [00:03:27] And that makes sense, right? It's Google using Google stuff. Um, again, like I said, I have no real recollection of, of this, and I think most people won't be. That's because most people really didn't use, didn't really use Google. They didn't use Bard at the time. Uh, you know, I did a little search and apparently chat GPT had 10 times the traffic then. So, you know, it wasn't the default tool for most people. Um, and that name actually only lasted a year and then went to Gemini. But suffice to say that the ability to have AI in your started for a subset of people that would that used Google, uh, in 2023. But true mainstream AI is in my email. That time didn't really come about until last year. [00:04:13] Um, and that's kind of the things that got me really interested in thinking about starting this podcast. I was just seeing, you know, again, I'm in the early adopter crowd and just seeing, you know, my guy K really using it, and I was like, what is going on here? [00:04:24] Um, but, uh, in 2025, there, I'm gonna back it up. In 2025, Claude released MCP. Right. MCP stands for Model Context Protocol, which is just how AI connects to software. [00:04:39] Uh, and that was actually, I got my dates wrong there. That was in 2024. So the end of 2024, November 2024, Claude releases. And by scroll, I mean anthropic releases MCP, which is model model context protocol, which is how AI connects to software. [00:04:57] If you want, check out episode 51. Um, WTF is MCP? What the is MCP? Um, and I, I did a whole episode on MCPS if you're curious. I will link that in the show notes. But MCP comes out in 2024, and by June of 2025, both ChatGPT and Claude were able to access Google Workspace and paying users had the option now to grant robots to grant the robots access to their tools. So if we zoom out, my present year, 2026 has really been the year of agentic AI, right? [00:05:35] But developers and other early adopters, early, you know, early, you know, AI early adopters and, and Vibe coders, they were getting a taste of this and they're getting a taste of agentic AI in 2025. Right? Uh, agent. Agentic AI, meaning AI that actually does stuff. It doesn't just answer questions. [00:05:55] So to that end, I do believe that the release of Claude code in June of 2025 definitely contributed to the willingness of at least the demographic like Vibe coders and the early adopters to grant AI access to their workspaces. Right? Because they're having firsthand experiences with cloud code and the capabilities. I mean like, yo, this thing can do stuff. [00:06:16] What else can it do? I want it to do more. The benefit outweighs the risk. I see the possibility. [00:06:23] Right. Again, not speaking for everybody. And I am privy to a, you know, a more siloed demographic, a more niche demographic as I was, you know, in the Vibe coding space and very much in the early adopter space. [00:06:35] Um, but you know, still as an outsider, which I think gave me a little bit of a different viewpoint because I think that people that are truly developers perhaps are m more willing because they like understand what's going on. Whereas I was like, I don't really know what this is, so maybe I'm going to just kind of pump the brakes, right? But if we fast forward From June of 2025 to the beginning of this year, January of 2026, Claude Co Work gets launched. And that brought with it the, the reality of, of agents for the average Jane and the average Joe. And that definitely, in my opinion, pushed along, you know, the early adopters within the general population, their willingness to grant AI access to things like Google Workspace. Right? So we had Vibe coders and coders and you know, the true early adopters using Claude code and being like, okay, I'm willing to let the, the robots be in my. Because I see the benefit. [00:07:29] Then Claude co work comes along and it's, it's something that can be used by the average person. They don't have to be inside of terminal. And you know, you know, Claude did come out, or excuse me, Anthropic did come out with the um, desktop app, but before then, and even with that coming out, you have to install that. And so there's friction points, right? But cloud cowork comes about and yes, you need the desktop app, uh, to do it, to use it, but it's just way less daunting than cloud code. And it's agentic, right? It's doing things for you. And you're like, yeah, I'm going to let it be in my Google sheets and in my Google presentation. What is, what are they called? Google Presentation, Google Slides, Google, whatever it's called. [00:08:13] It made sense, right? Because it was. People could start to see the benefit. [00:08:18] But I don't want to act like it was just mass adoption because, you know, people are still very. Trepidation. Trepidatious about it. Right? But if we zoom out and we speak to the sentiment held by the general public and we start in 2025, there was that hesitation surrounding granting AI access to these sensitive workspaces. [00:08:36] I said this, you know, I must privy to some of these siloed spaces and I saw things on Reddit and newsletters that I subscribed to and just talking to my friends and people in the space. And, um, even early adopters, they were, they were cautious, right? Yes, of course there were the, the YOLO folks who live by the, the brochacho mantra of like, move fast and break things. But for the most part, early adopters were cautious and the general public was, you know, that's going to be a no for me dog. [00:09:05] But within the course of a year. Within the span of a year, right? That was 2025. [00:09:11] We're in 2026, October. And I don't have any hard data on this and I cannot speak to everyone's experience, but between June of 2025 and present day October, again, October 26th, uh, my own personal willingness to grant AI, you know, especially Claude, access to my workspace and other apps has definitely increased. I was so hesitant at first and you know, right now I had. It has limited access to Drive. It has access to PowerPoint, access to Excel. Yes, I'm old school. I use those things. It has access to my web browser via Claude in Chrome. It has access to my calendar via Granola. It has access to podcast transcripts via, uh, descript Connector, and probably has access to other things. Those are the things I could think of off the top of my head. It does not have access to my email or like my bank account. [00:09:58] But I, I can't help but if wonder if it's just a matter of time, especially with things like the email, though I don't have any, like, use case for it. Um, but in general, the things that it's not that it doesn't have access to, I can't help but wonder if it's just a matter of time before I granted access. I don't care. Which brings me to the biggest concern I have, which honestly is not privacy. It's prompt injection. So I spoke about this in the MCP episode, but prompt injection is when someone hides instructions inside of, uh, content, like an email or a document or even a webpage. And those instructions are meant for AI, not for you. You cannot see them, but AI can see it. And so AI sees it, it reads it, it follows those instructions without you even realizing it. And then bad things can happen, right? So, for example, you know, you grant AI access to your inbox. You want it to, like, read, write, and take care of emails and send things out. [00:10:52] So let's say a shady character sends you an email that has a prompt injection. [00:10:57] Your AI reads that email, it reads those hidden instructions, and it carries out those hidden instructions, which maybe it says to send that person confidential information that, you know, your AI also has access to because you've given it access to everything. [00:11:12] Right? So that would require that. That scenario would require AI to be able to both read and write and act. Right? Uh, which brings me to my next point, which is a framework that you can use, because like I said earlier, this, this episode's more about just offering up some thoughts, uh, and some suggestions, some questions. [00:11:31] Um, so next point, here is a framework that you can use, slash, some questions you can ask yourself when you're thinking about what should my AI have access to? Right? So two questions for. And that you should think about before any connection that you make. [00:11:47] Read, write, or act. Right? Which of those do you want it to be able to do, AKA can it see stuff that's reading? Can it change stuff that's writing? Can it act? Can it actually do stuff? [00:11:58] Uh, can it actually send an email? Difference between writing an email and then sending an email. Two different things. [00:12:03] And then the second question is mine or other people's, and that relates to the data. [00:12:08] So if we're looking at low stakes here, that would be read only. And it's your own stuff. So, like, connected to, like my podcast, to be honest, completely honest, by my podcast transcripts. It's just me talking. It's not a guest transcripts or anything like that. It's literally just my own stuff. It can't do anything, right? It can't, like, send anything. [00:12:26] It can, in this case only, it can read. [00:12:29] And, uh, that's it. It can't change anything inside of it. It's just copying it and then putting it into, um, you know, making these, giving me a response inside of the desktop app. [00:12:41] High stakes on the other end of the spectrum here would be that it can act and it touches other people's data so it can actually send things to other people. That'd be an example. I can send an email. It's acting and it's sending it. It's reading people, reading emails and sending things to other people, or it's reading other people's data. Um, so I cannot tell you. [00:13:03] Um, let me back it up. When I see reading other people's data that could even include like any of your call transcripts, that is other people's data. Um, so I can't tell you what you should or should not do. [00:13:16] You know, what you should choose or should not choose or what's right for you regarding access. [00:13:20] Uh, and honestly, my own answers to those previous questions, they will vary depending on the task and depending on the specific platform. And honestly, depending on the specific possible, you know, possible slash, potential benefit to be had. [00:13:37] Right. What I do want to speak to quickly that I just alluded to a little bit, there are privacy concerns, right? Uh, data and privacy. We'll get to privacy first. When it comes to privacy, honestly, I don't care. [00:13:48] I don't care. It's my own personal feeling about this. I feel like I get a fucking email every day that some important information from some bank I use or some shit that I have has possibly been compromised. Like every single password has been compromised, always, ever. And I'm like, what? All the time in these emails, Right? Every app on my phone seems to be tracking everything, even after I turn off tracking. [00:14:12] But you, you just think something and you get an ad for it on Instagram, there are cameras literally everywhere. I live in LA County. Like, I'm not living off the grid or anything like that shit. I have Global Entry. I don't know if you ever, guys, if you folks have ever used Global Entry at this point. You just walk through the thing and I'm like, well, that means my face is just like scanned into something somewhere. But, like, it's faster the benefit and outweighs, uh, the risk. So I am not under the impression that I actually have any privacy anymore. Ah. And so I'm not really concerned about that. [00:14:41] Uh, and that thus it's not a top factor when I'm deciding whether or not AI should have access to something. [00:14:49] Again, I'm most Concerned with prompt injections and bad humans possibly hacking me. Uh, so the problem injections really would require that AI can also act, not, uh, just be able to read or write. Right. It has to be able to do something, send something, but make something. Right. But the next part here is training. I think some people are concerned about that and any AI training on your data. And honestly, for me it's more that I just am like you guys, right? Like, I don't want them training on my for free. And technically it's not even free because I'm paying them. So I'm like, wait, I'm paying you and you're using my to like, make this better? Maybe, but probably not immediately. No. So I turn that off out of principle because I'm like, scared or because I care that my shit's out there. I'm just like, you know, that's, that's literally why. [00:15:35] So a question that is worth sitting with, and I said earlier, we talked about privacy and then we'll talk about data. We're talking about data. Now. A question that's worth sitting with is what about access to your client's data, right? Or people that you're working with? What about access to someone else's data? [00:15:49] So for example, transcripts have clients information, right? And their data. [00:15:54] You may not care about your data being out there, but what about your clients, right? This, this decision, this question, this answer is going to be personal. [00:16:01] This is going to be industry dependent. In my opinion. Some things like, may feel more confidential. [00:16:06] Um, and that's still, I think there's even a personal aspect to it and how people rate it and how, how personal and private they write that data. [00:16:16] But ultimately it's up to you how you want to handle it. I'm definitely seeing more coming out in the legal side and disclaimer side of things. But of course I'm seeing it come from, from like lawyers. And I'm like, are you just saying this because you want money? Like, can I just use AI to do this? But either way I am seeing talk about it and like, are people, are they talking about it because they need money and they want to make money? Probably. But, um, definitely seeing more about it. Um, worth noting. Yeah. You know, you can anonymize things before you enter them into AI. So if you're like, have a bunch of transcripts that get dumped into AI and you use it to like pull pain, uh, points or use it for, you know, to create marketing copy. Yes, you could totally anonymize it first before you Put it in. Um, but that's not always the case where you can anonymize things like something like a note taker, like the person's name is there. [00:17:08] Um, but it's just that name. It's not all there for I, you know, these are just things that I want to throw out there. [00:17:14] All right, I will say another thing to throw out there is that if I flip it as a client, you know, I would love it if my therapist used AI if she felt that it helps her be more present during sessions and made her post session work easier. I'm like all four of them. Yeah, absolutely. [00:17:32] I think that there is, you know, perhaps a muddier, murkier grayer area when it comes to like physiological health and health care. Mainly because you know, I just feel like in the, you know, same thing can happen with mental health though, you know, in the wrong hands then people like discriminate you against you because your medical information's out there. And then they're like, you can't get a job. I don't know, like it doesn't really apply to me. [00:18:01] Uh, or, or your insurance cost more like, you know, the people just use it in the worst kind of way. And so again this episode isn't what information should these companies have access to and should they store? I'm like don't store anything. And the solution there is, you know, use a local model. [00:18:17] Um, but in terms of you granting access, this is again, I just want to throw, put these questions out there and in this case it's just thinking about what about your clients data. [00:18:30] All right, so this is a uh, a good segue. I think I'm just going to make it a segue for my, my 30,000 foot belief on this, which is that ultimately AI access does and will continue to come down to benefit versus risk. Right. Every day we see AI getting more and more capable. At the time that I'm, you know, recording this opus 5.5 is the default model. They just, they come out with new things every two seconds. All right, so AI is becoming more capable, more people using it. As more people are using it, it becomes easier for folks to see and experience firsthand the benefit that that can be had. [00:19:12] And when that very real, very tangible benefit starts outweighing the risk of something possibly happening, you know, related to data compromise or workspace mistake, I think folks become more willing to grant AI access to their stuff. I uh, and yes, I'm very much speaking from personal experience here and just my own, my own process and evolution with Things. Um, even with something as simple as within. Within a session, changing the permissions from. In Claude from. Oh, allow. Wants to always allow. [00:19:45] So this is if I'm using any. The agentic side of things. If I'm m. Vibe coding, I'm building something. [00:19:51] Um, but mainly I just did that. This. Just did this with, uh. And you'll see, you'll. I'm going to talk about this in a few seconds. About, um, connecting Descript. Connecting AI. Wow. Connecting Claude to Descript. And I just made it always allow. I was just like, I can't be clicking this thing each time. Like, it's fine. I see what it can do. I'm fine with it having access to this. And it just has. It's not like it can actually, like, make something or delete anything. So it's fine. [00:20:17] Um, but you know, again, when the tangible, very real benefit outweighs the risk of something possibly happening, right. When there's the current benefit, always a possible risk, then, yeah, people are going to be more willing to grant AI access to their stuff. All right, so what should your AI access have Act. Wow. What should your AI have access to? That is up to you. Right. I think it's prudent to take the read, write, act approach, um, that framework that I. That I laid out earlier, and not just give it no carte blanche access from the jump. [00:20:52] But as always, it's up to you. [00:20:56] All right, let's go into the last section here. Wrap this up. So how I use AI this week, each episode. If you're new here, welcome. I share a quick example of how I used AI that week. This week. Like I alluded to a little earlier, I connected Claude to Descript. Descript is one of the video editing software tools that I used. Uh, and I had it go through the transcripts from my podcast. Not this podcast, my other podcast, my showing them. Like, I have 700, like 30 episodes. I had to go through the. The transcripts that are uploaded to it, um, that have already been uploaded to it. [00:21:32] And I had it pull quotes that I could use for social media. [00:21:35] So as we know, the better the input, the better the output. That means that first what I actually had Claude do was go and look at my Instagram, and I had to go look at the posts from the past, like, month or so, whatever it could see. And I was like, look at the quotes. Get a feel for what I'm. My, my, my voice, my vibe, um, what, you know, the type of quotes that I choose and that I put on this page. [00:21:58] Um, and this way could use that to go and pull more that are the same vibe, right? So I did have to go back and forth with it a few times, a handful of times, something crazy, uh, to get the, quote, suggestions that I was happy with. And one of the. The changes that actually Claude. Claude suggested, and then I was like, oh, yeah, let's do that. Was just allowing it to pull copy from across multiple sentences and multiple paragraphs. Because, you know, this is pulling from a transcript. And oftentimes when you pull from a transcript or you're speaking, the point comes across and it gets delivered across sentences. It doesn't just happen in one punchy sentence. So I was like, yeah, go ahead. Like, pull from multiple parts of it and synthesize. And so with doing that, uh, things were much better. [00:22:44] So each time I had the. I asked Claude to do this, I just did it like, you know, two or three times. Maybe, uh, like three or four times. Uh, each time I did it, Claude gave me like 30 quotes. And they're definitely not all usable. But overall, I am super happy with, with what it turned up and what it came up with. And, uh, I will definitely, uh, be returning to the proverbial. Well, I, um. All right, that. I'm looking at the clock, I'm looking at the outline. That is all for today. Hopefully you found this episode helpful. If you did consider leaving a little rating. Rating or review would love that. Don't forget, I have a companion newsletter and blog called the Curious Companion that drops every Thursday. And it is basically my basic, I mean exactly, the PODC podcast episode in text format. So if you prefer to read or you just want a written record of the things, join the newsletter or check out the blog. You can head to prompting Curiosity. Wow, swallowed those words. You can head to prompting curiosity.com forward slash newsletter or prompting curiosity.com forward slash blog. Or keep it simple and just check out the links in the show. Notes us, as always, endlessly, endlessly, endlessly appreciative for every single one of you. Until we chat again next Thursday, stay curious.