Everything from our session, and where to start.

AI FUNDAMENTALS — SESSION ONE
Flintco × Texas AI Lab · September 17, 2026 · 56 minutes
Led by TJ Larkin. Recording: https://fathom.video/share/inPeqYjfPCtyjouciVxWaZdMxpuiMUH4

Note: this is the meeting transcript, lightly cleaned of duplicate passages from the
automatic transcription. Speaker names and timestamps are unchanged.

---

[0:07] Tony DeStefano: TJ, I apologize. I just looked at the invite and I sent it to my team and I didn't send it to you. So that's on me.

[0:19] TJ Larkin: Well, I'm here to tell you all about that, all the failures of AI, but there's a lot of great stuff too. Hopefully I can help you guys a bunch here.

[0:46] Tony DeStefano: Everybody, TJ is a consultant that I was looking for training to do, and found him actually in the Austin area. He's going to spend half a day with us when we get together here next month. Part of his training program was a pre-work call and a post-work call. So this is kind of an introductory call. I wanted him to introduce what's going on with AI, some of the trends and things which change daily. And I've told him, when we get together in person, I'd like us to see if there are ways and things that we can use as it relates to AI to automate some of the lesser value, non-value-added tasks that we do a lot in our roles — whether that's sourcing candidates, document creation, the filing, the approval workflow processes — so we could spend more time doing value-added tasks in our role. So this kickoff call is going to be more or less an intro. I told him, let's start really slow. Some of this may be redundant for those of us who've dabbled in the AI space, but we have a broad group. I want us to start as broad as possible and narrow it in.

[2:14] TJ Larkin: One of the challenges is knowing where everybody's at. Some stuff might be a little basic, some not, but I'll do my best. Feel free to jump in anytime with questions.

So, I'm TJ, and I was an insurance agent for 20 years up until two years ago. I have no tech background at all. I've never touched a line of code — I do now, but that's only because AI does it for me. Even though I know nothing about HR, I'd consider myself more like you guys: a professional services type of background. So keep that in mind when you see everything here: I don't understand technology really at all, or at least I didn't when I started doing this. There is no background needed to do almost anything you're going to see.

First off, this system that you're looking at, I built this yesterday for you guys for this presentation, the whole thing. This was me just dictating to AI what I wanted, and it built this for me last night while I was in the garage working out. Now, I did have a little bit of a frame here — I built something similar before that I told it to replicate, so I may have had a slight advantage. But this is what's possible today. Everything you're seeing here I made yesterday, just using my voice. I didn't type anything that you're looking at.

That's one of the big things I want to go over with you. The number one thing, for those of you who are just trying to get started: when you don't know how to do something with AI, the person you ask is AI. "How can you help me? I have this workflow that's really annoying. I've seen people make documents like this. How would I do it?" Ask it to help you use itself. It sounds obvious as I'm saying it, but you probably don't do it, and you should. Say: I want to accomplish this goal, how can you help me? You don't have to go to it and say, please go do step one, two, three. You just ask it to help you, even if you don't know how it's going to help you.

This is probably the most important takeaway, because you're going to leave this call and think, I saw a lot of interesting things, but what do I actually do? I don't remember the specifics. Just ask it. Whatever you remember, just say, "this guy said this — what can I do here?" And it'll help you.

The other one: we're all human, we get in our comfort zone, we do what we're used to. Hopefully I can show you some things today you'll be a little bit excited about, even the smaller single-step things. Just try to lean on AI a little bit more each day to see what it can do. When you take one baby step, that's good. Don't stop there. The next day, take another baby step. You'll keep seeing more and more improvement. It's that whole concept of a rabbit hole: once you start going, you go deeper and deeper and you start to see what's possible.

[5:53] Tony DeStefano: And leaning into that, if you're using one platform — whether it's ChatGPT or Claude or whatever — as you spend more time in it, they learn from you and kind of carry you even ahead of where you think you might be going.

[6:15] TJ Larkin: Yes, and I'm going to get into that on memory, or really context windows. It will get smarter about you, and there are better ways to do that, which I'll go over. That's when things get nuts: when it immediately knows what it did for you last week and you didn't even remind it.

Quick basics. You hear the term tokens a lot. A token is actually three quarters of a word. Just think of a token as a word. It's all trained on the entire internet, basically, of text. They're now doing training with video and audio and images, but think about it as AI's read the entire internet — that's its intelligence, and now it's about how they leverage it.

Technically what these LLMs are doing is predicting the next word. You've probably heard that before. It can be hard to grasp when you see how much it can accomplish, how fast. Just remember it's doing the next word a million times faster than your brain can work.

Context window: you probably won't deal with this a ton in the short term, but longer term this is one of the biggest problems in AI — how much context can it have. I'll show you Grokbot at the end, which is one of the newest things. I had a problem with it where it said, you've talked to me enough now, my context is gone, you have to start a new agent. It can't just know everything about you forever. Most of them have about a million-word context window. If you go beyond that, it has too much, it can't handle it. It can only keep so much of the context in its brain before it gets confused. This is possibly the biggest problem in AI. If they ever solve it, it's going to be insane.

AI is wrong sometimes — everybody knows this. I'd be less worried about hallucinations, though they're still a thing. The problem is more that it's very confident in whatever it decides. This is where the human piece comes in. You're the HR experts. You don't want to let it overrule you on things that you know, because sometimes it will go in a wrong direction and be like, oh no, this makes total sense. And you say, no, I'm in the industry, that's not the right way to do it. And then it'll go, oh yeah, you're right, I guess your way is better. So you have to be prepared for that.

The other key thing, for those of you who've mostly just done ChatGPT — rewrite my email, search for things — is that AI now has access to tools. That's where things like API keys and MCPs come in, which really just means two tools talking to each other. It's like a bridge: how do you get Salesforce to talk to Canva? You need a bridge in the middle. These AIs can access those tools, which means they can take actions on different platforms.

[9:55] Tony DeStefano: Just one comment for our group. We have put together an AI governance council, which I've been asked to be on, and the "confidently wrong" topic is something we were talking about yesterday in our meeting. We talk about employees wanting to use all these AI tools and software platforms that are sprouting up all over. One of the things I think you'll see is we'll be a little bit more open to people using some of those tools if they're using them in the field they're experts on. So for example, if we're using an AI tool to review legal documents and contracts, our team is not the one that's going to get the green light to pilot that — our lawyers are. For us, I think we're going to see the company move in the direction of, here's this thing that we could do for development planning, and that will be something we can look at since we've got some of that knowledge. The tools being confidently wrong, we can kind of override it and make sure there are boundaries and parameters around them.

[11:06] TJ Larkin: Yes. That's something I know is super important in a space like yours. When you're more in the entrepreneurial world, you just do whatever you want. But it is important to remember all that security stuff. The confidently wrong thing is so important — to question every answer you hear.

You just brought something up for me, Tony. It's not on my sheet. This is one of the most valuable things that I do. I do it almost every time now. I do a more complicated version where I'll have Claude review ChatGPT's work, or ChatGPT review Claude's work. That adds complexity, but here's the simple version. When you get an output that matters in some way, that's not simple, ask it to review itself. Just say: I want you to score yourself, zero out of a hundred. Give yourself ten topics and give yourself a score on each. And then for anything that isn't a ten out of ten, list out what you missed.

You will be shocked. It will be a six to eight out of ten every single time. It will list out all the things it did wrong. And then you just say, okay, fix those things. And you will get a massively better output. I know it sounds silly — why didn't it just do it the first time? That's just not how it works. So: ask it to review its work, ask it to score itself, anything that's not a ten out of ten give me the reason, and then redo it to solve those problems. Super powerful.

As far as which AI you should use: I know Tony said you have ChatGPT. Personally I like ChatGPT the most right now. The media is always a little bit behind; Claude is the one they talk about now. Claude is still great, don't get me wrong, but a lot of people are liking ChatGPT more now. Honestly, for you guys, it doesn't matter. I'd put ChatGPT and Claude as the two top dogs that are the leaders in almost everything. Gemini has some advantages, which is Google's product. And Copilot — admittedly, I don't know much about it. I know a lot of companies use it, but I can't really say too much. It doesn't get discussed very often in the broader AI world. I try not to be too negative.

Here's one of the most important things to think about as far as where you want to go, and this is what Tony already said. Most knowledge workers are starting to understand this: moving from asking ChatGPT for things — write me this email, write me this document — to having it take actions for you live. And then the best version, which is what Tony mentioned, is automating things: hey, every Thursday, take my reports and push them into the CRM.

[14:35] TJ Larkin: Automation runs on a trigger. A trigger can be a date — on Thursdays — or it could be, every time this happens, take this action. To clarify, for any of you who've heard of a tool called Zapier, which has been around for like 15 years: Zapier has been doing this forever. The difference is that up until a year or two ago you had to know how to use it, you had to go click all these random buttons, and it was annoying and hard. Everything has changed. You just use your voice now, or a keyboard if you want, but I like to use my voice. You can do all of the automating by just describing it. You don't have to go into Zapier and do all this tech stuff. That's the real game changer.

[15:36] Tony DeStefano: I think we're all conditioned to think about how do I solve this problem, or how do I complete this task. And now you've got this whole new tool that, if we think about how we engage with it daily, is really going to provide a lot of strong, powerful resource for us to do things more efficiently, free up time, and do some things smarter and better. So the ask, act, automate — I'm still heavy in the ask. Hey, do this for me, take a look at this report, do this letter. And now we're going to get to a point where we've got things set up that it's going to act and come to me and say, hey, I've got suggestions, where do you think we should go from here. Which is really exciting.

[16:40] TJ Larkin: It is crazy exciting. The amount of things you can ideate — every day I come up with new things. Once you start to understand it and grasp what's possible, you'll rethink everything you're doing. Why am I doing this? Why don't I have the agent do this for me? Why don't I have a way to automate this? Why don't I connect this thing to that thing?

Here's another super valuable thing. Understand in general with AI, it can accept any type of content at this point. I do this easily ten times a day. I'm on Twitter, there's a lot of AI people on Twitter. Whenever I see somebody post something interesting, I just copy the link and drop it in and I say, "how can I use this?" That's it. Literally one sentence. And it'll give me a whole thing: oh, you can do this, this, and this.

You can drop images. That's one I do probably fifty times a day, because I screenshot things I'm building and I say, look at this problem, what do we do? PDFs, photos, copy-pasted text. Knowing that you don't have to type things — give it information from any sort of source. Very, very valuable.

Let me show you something. I'm telling Gemini right here to interview me, to create a document for me. That's literally all I did with my voice. I said: I'm an HR specialist, I want you to have all the information about me to leverage in perpetuity in the future. So it is creating an interview that I think all of you should do yourselves. Answer all the questions, and then when you're done, it will give you a document — an AI context bible about yourself, your company, what you do, all that kind of stuff.

Then that's the document you want to store anywhere that you're having AI work for you, because now it will have — this is the better memory thing I was talking about. Instead of just hoping it remembers you, why don't you create a document about everything about you and what you do, and then store it in the system so it always knows? All I did here was tell it to come up with the interview. I'm not an HR person; you can look at this and figure out if it's good or not. What are the three to five tasks I do every day, what HRIS, et cetera. So every time you're using any AI tool, it will know everything about you.

I dropped a YouTube video in here. I do this one a lot with YouTube videos: give me all the takeaways, I don't have time to watch it, give me the top three takeaways. Only Gemini can do this, by the way, because Google owns YouTube. Other tools can do it with workarounds, but with Gemini you can literally drop a YouTube link and say, give me the summary.

[19:40] TJ Larkin: Don't just tell it what to do — let it interview you. There are two versions of this idea. The first is what I just showed you: a permanent interview, one time, maybe you update it every once in a while — a one-time document for you to do. The other thing you can do, and should do, is when you want a task that's kind of important, ask it to ask you questions.

If you know, hey, I need to create this document to do this thing, you could just go in and say, "create this document and do this thing." Or you could say, "create the document, do the thing, and then ask me five questions about this to make sure you fully understand the task, and then do it."

This is why you get so many of those confidently wrong answers: if you only give it a little bit of information, it's kind of sort of guessing. By saying "ask me questions," you're giving it permission to clarify. And that helps a ton.

Think of AI as a confident intern who doesn't recognize he should ask you questions. If you went to an intern and said, hey, I need you to go do this thing, he's just going to go, yep, I got it, I'm gone. And you're like, yeah, but I only gave you one of the ten things that need to be done. It's not going to ask for the other nine. It's just going to go. So you want to set the frame, and the best way to do that is have it ask you questions.

[22:40] TJ Larkin: This one is related to the document. In ChatGPT — although, speaking of things changing, Claude may have just solved this problem on my slide last night, after I made it. If you've heard of Claude Cowork: what Claude Cowork was until last night is it would use your files on your hard drive.

If you just go to Claude regular, or ChatGPT, it's not referencing anything outside of its own system. Claude Cowork allowed it to access your hard drive, your files. That's where you would save that interview file, that document about you. And that's where you can save all kinds of things. The bible of yourself is just one. You can add in your rule books. You can add in examples. You can add in a company document. Tony, I assume you have one, or should create one — that same concept, everything about Flintco. And that should be in everybody's file as well. So they have everything about the company, everything about them in their role, and the rules and other documents from the company, all in that folder. So then every AI references it going forward.

[23:22] Tony DeStefano: So TJ, when you do that, do you just drag and drop it in? Using ChatGPT as an example, you just drop it in there? Or is there a, because I think Claude has, what, relics or something like that?

[23:37] TJ Larkin: You're probably thinking of artifacts. There are two ways to do this, and I know I sometimes jump to the further one, so roll me back if you need to.

The simple way: ChatGPT has projects. You create a project, and then you can say, do you want the memory to be default or in the project. And then there's a place for sources. This is where you add documents. So this is the version where it's all inside of ChatGPT, not on your hard drive. It's the easier version. It's not quite as good, but it's still ten times better than if you just go to a blank chat, which is how ChatGPT has been for three years, where you're just hoping it's referencing its memory. And hope is not a good strategy. A blank chat is good for casual chat, but that's it.

[24:46] Sean Williams: TJ, I thought you set it up under your profile there in the bottom left corner. Your name is TJ Larkin. If you click on that, personalization, isn't that where you set up your profile? Where you give it things about you?

[25:07] TJ Larkin: You can, yeah. Right here, custom instructions. This is where you could put in, I'm TJ, I'm an HR expert, and you could do that. I don't do this because I have it all on my hard drive and it just reads my hard drive. I don't even use ChatGPT, I use Codex. Codex is the — people think of it as coding, but remember, coding is now English. I know the word code is scary as hell. It used to scare me, I never did it. But code is now just English. So you don't have to be a coder to know how to do it. But yes, Sean, you can just do that.

I do think projects are key. Here are the instructions, and here are the sources. This is my media company that I do here in Austin. Here are our logos, our launch checklist, our SEO strategy, our admin guide, our runbook, our content, the voice of our brand. I did this so that my wife and I could share it. This is how it would look if you just did it in ChatGPT. And Claude is basically the same thing.

[26:27] Tony DeStefano: Just for reference, we started using Claude. Claude was doing some of the cowork and had access to some files, but then — right, wrong, or different — we noticed Claude was starting to look at different files that it didn't have access to in our network, and so they shut it down there. Now, I'm sure that's fixed at this point, but as a general point, I think we would probably do it by project, where we'd put in, like, I've got one that's work-related presentations, and we're doing our business plan, so I've got a 2031 plan, and in there is where I would set up those sources specific to whatever that topic is.

[27:09] TJ Larkin: These are the things that ChatGPT and Claude are going all in on — B2B, enterprise, for people like you, to streamline this stuff. This is going to change so fast to make those problems easier. This is just really good for you to understand as humans, even if it changes a little for the company.

Here's my favorite slide, because Wispr Flow is just one tool — there are a bunch now, Wispr Flow is the biggest — but this is the tool why I don't type anymore. You just push a couple keys on your keyboard, or I have it on my mouse, and then you can just talk. I just hit it on my mouse right now and I'm just speaking out loud.

Let me show you. I'm going to say: I want to do bullet points. Number one is very important. Number two is kind of important. Number three is not. Lay this out for me.

See, with these tools — because of AI, it does the bullet points. When you're speaking, it'll take out certain words, it'll even go back for you. If you say, wait, no, not that, it'll actually delete that part of what you just said. Because of this, I don't type more than one sentence. I feel like this increases productivity a ton.

There are two benefits. Number one, it's easier and it saves time. The second is a huge benefit with AI: you have to reframe how you think about search, or Google. Google was all about keywords. You didn't want to type too much because you might mess up the keyword you were trying to attract. That's all gone now. The more context you give an AI, the better.

When you're typing, your brain is faster than your hands, so you're editing. With AI, you don't need to edit. You should just brain dump — whatever's on top of your mind, you just drop it. And the best way to do that is with your voice. So you're actually helping the AI by using voice, because you're giving it more context. Two good reasons to do it. Now, if you're all in offices next to each other, there's an awkward angle there, but speaking in general, using voice is better.

And then ChatGPT voice — I'd recommend everybody play with it. Use the voice system. It changed about a month ago. It is like that movie Her now. We have crossed the Rubicon. I go running for an hour and do brainstorming sessions with it that are just absolutely insane. You have to try it. If you want to believe how good AI is in five minutes, talk to ChatGPT voice and you'll be blown away.

[31:00] TJ Larkin: Deep research. This is a good point, and it doesn't have to be for deep research, it's for anything: when you want something, if it's a little bit complicated, don't ask directly for it. Ask for a prompt.

The example I did for you: I want to do a deep research report. I went to ChatGPT and I said, I want you to create me a valuable deep research report on the construction industry that Flintco works in. Give me a prompt that I can give to Gemini to do the deep research. So this is step two — it gave me this prompt. Very long, very complicated. And then this was the output. It's about your company and your industry, and it's like 40 pages. Here are all the sites that it researched to create this report.

The way I like to think about it: these consulting companies, McKinsey, Bain, this is what they used to deliver for, whatever, twenty thousand dollars. And now you can do it in ten minutes for free.

And even better — you see this button, "create"? You can turn it into a webpage, an infographic, a quiz. I already turned it into a webpage. This is the same report you just saw, basically as an HTML website. You can even ask AI about it or do a simulation. I literally hit that button. Here's the full report as a document. Sometimes these reports in Word documents look amazing; this one doesn't, but you can see it's still got all this stuff.

I also went to Gemini Notebook, which used to be called NotebookLM, and I dropped the report in and said, make me a video about this report. [video plays] It literally made a ten-minute video about that report. This took two minutes of my time. I copy-pasted it in, told it to make me a video, that's it. And there's other stuff you can do — you can make it a podcast, you can make it a slide deck.

[35:40] Sean Williams: TJ, for Gemini, is there a fee associated? Are you paying for a license with that?

[35:45] TJ Larkin: Good question. Google gives away a lot because they have all the money from their advertising business that the other two don't. I have Google Workspace, so I do technically have a paid account. But you can get a lot from it. There's also Google AI Studio, which I'd recommend if anybody wants to try some stuff for free. It does a lot where you can build websites — it's kind of their playground to just do random stuff.

Real quick, images. I think you know you can make AI images. Everybody knows now. I built an AI image generator into this platform for you guys with no effort. I'm honestly a little shocked that it did this. I just told it, make me an image generator. I had to give it an API key because it costs money to run the images. Other than that, it did this on its own. So not only are images easy, building applications to make images is easy. Even I was a little surprised by this one.

ChatGPT is by far the best for image generation. They just came out with their new model two weeks ago. It's just insane what they can do. And that's a good one, too, for asking for a prompt instead of what you want. If you want a photo of you looking like you were dancing in the 80s with Abraham Lincoln, instead of just saying that, go ask for a prompt that will get you that, and then put that in ChatGPT.

[36:40] TJ Larkin: Now here's where the gold comes in. We're getting into the workflows, the important stuff. And Tony, you and I should talk before we come on exactly what systems you want, because this is where security becomes real — what can you connect, what tools can you connect to your system.

Again, don't worry about APIs and MCPs. It sounds confusing. It just means connect two apps. ChatGPT and your email. ChatGPT and Canva.

This is where things get really exciting for AI to help you way more than just plain chat. You sit in the middle and then you connect all of these different tools to your system. For you, I think a ton of it's going to be Google Drive type of stuff, probably Excel, Gmail, calendars. So many tools now, especially enterprise-level ones, almost all have connections. It's rare to see one that doesn't. So you can think about your ChatGPT as being able to talk to almost any other tool you use. It just becomes a security question at that point.

I built my own software to do newsletters, because that's what I do a lot of here in Austin. ChatGPT can use my software directly. I told it to make it an API. It took 20 minutes, and now I can ask Claude to use my software for me.

[39:20] TJ Larkin: This is probably one of the biggest takeaways I want you to see, and I'll do it live for you. This has been an absolute game changer, especially since they fixed another problem in the last two months: the Chrome extension. Does everybody know what a Chrome extension is?

So you've got your Chrome extensions up here. I use this all the time now to either research what's on my screen or ask it to do things for me. That's where it gets really exciting. Let's see.

[typing] Codex, I want to apply for this role. Can you please help me start filling out this form? You have my authorization, go ahead and get started.

I'm a little concerned it won't, because it's asking for personal info. Sometimes that might get a little weird, but there's a bunch of info here.

Watching AI take over your machine and take action for you on the website you're on — I guess I'm kind of desensitized to it now, but I remember the first couple of times it was just like, oh my God, this is incredible that you can do this. I do this easily ten times a day with all kinds of things.

Oh, crap. You saw the mouse come up and then it stopped. Let me see. [retries] Just fill out the best parts of it you can, and then I will fill out the rest.

Has anybody done this before, seen AI take over their machine? Or in this case, it's just taking over the website — it's not taking over my whole machine, which I'll get into next. Sean, you have? Have you done this, Tony?

[Tony: I have not, no.]

I landed on the wrong form, but I promise you it works pretty well. I'll have a better one to show you when we're there in person. It really is incredible how you can get it to just do anything, research anything.

What I was going to say is, when you have those long websites and you're like, I don't want to read all this — you don't have to copy and paste the link and go back to ChatGPT. You could just open this extension and ask it. Summarize this for me. What's the key takeaway?

And then the other one is just taking over your computer. I won't mess with this right now. This one can actually be annoying sometimes. I do it a lot where things will just happen on my computer, because an agent is building something for me. It can be a little distracting, but it's still cool to know that it can do it. So when you have an app on your computer, you can tell ChatGPT or Codex to go use it for you. For example, Photoshop. You can say, go use Photoshop and design me something, and let me watch you do it on my computer. It can do that.

[42:40] TJ Larkin: Slide decks. I guess I shouldn't say I don't make slide decks, because you're looking at one. But in general, if I was sending something to someone, I make it as a website. That way it just scrolls down instead of across.

This is basically an HTML file. Another way — and Tony, I'd be curious what your team would think about this — I know a lot of companies are trying to do this now: instead of sending a document to people, why don't you just convert it into this and have this be what people get emailed? Here's our daily update, or whatever it is. You can just turn it into an HTML file and send this instead of a document. And it can have tabs — this is one file and it has all the tabs, and you can do all kinds of stuff.

This is generally what I would do now if I was making something like a pitch deck or a presentation. I would just do this, or I would make it as a website and you scroll down instead. It's very easy. In the Google one it had a button for it. When you're in ChatGPT or Claude, you don't need a button. You can just ask for it. Just say, turn this into an HTML file, and it'll do it. Again, everything is just voice. It understands English. You just tell it what you want. Now, sometimes what you get will be bad, but then you just tell it to iterate on it.

And then one thing you could do, Tony, as a company, if you haven't already, is have a stock presentation style. You have a branded company look so it always looks good, so people aren't sending them out and they always look different because of their ChatGPT. You have the standard design. Everybody's could come out that way if you had one stock one in the project.

[43:49] Tony DeStefano: In fact, we just updated our brand standard. So all of the preformed templates I used to use, I've had to update because they've changed it. So it'd be easier if I could just drop it in there and go forward.

[44:02] TJ Larkin: That's it. Look what I did for mine. This is my brand. I built a website for my brand that has all the colors, font types, all the components. I don't even know what some of this stuff is, it just did it for me. So I turned it into a document. I want to just drop this link anytime I want to build something. I hope you'll have something like that, or you will soon.

[44:20] TJ Larkin: Here's another key point. You probably do a lot of Zoom calls like this and meetings in general. When you have a recorder — I'm using Fathom right now, Fathom, Granola, whatever — where the power of these agents comes in, and I know this is a little complicated to set up, but this is more to think about what's possible: this stuff gets documented on your computer, in your project, wherever. You can then tell agents after the fact to go use that information.

For example, I actually have this set up. I only did it last week. I have my agent going after my Zoom call. He'll do it after this one. He will come back to me in 30 minutes after this call and say, hey, here were the top takeaways from your call, here's what you need to follow up on, and here are the things I can do for you. Like, you mentioned you were going to send them this document. Do you want me to go write it for you and send it to them? Yes, go do that. And then it'll do it. I didn't have to ask it, because I set it up on a trigger to read the output and then take action based on the output.

I know you're not going to do this tomorrow, but this is what I mean about ideating on what's possible, and understanding why having these things saved somewhere — in a project, on your hard drive, wherever — matters, so that agents can take action on it.

Here's the memory concept again. Really just the project thing we talked about, having all the documents.

And then, meetings become material. This is huge. Even for me, for creating content — the agent goes behind me and says, oh TJ, this was a really cool idea you talked about, let's write a LinkedIn post for you about it. You want me to do that? Okay, great. You probably don't need that, but I'm sure there are other things you need that are more valuable that it could go in and do for you after it hears you discuss it.

[47:00] TJ Larkin: Here's another one that I assume you'd want: your own daily brief. I get a bunch of emails every day from my own system. One, I have my own personal newsletter about anything I care about. I care about AI, newsletters, marketing. So I get an email about all those topics every day. It's a custom, personal newsletter.

I also get a daily briefing: hey TJ, here are the things you did yesterday, here's what's due, here's what I can do as an agent for you. It's taking everything it learned from me, from my calls, from my chats with AI, and it gives me a daily summary of the past day and what I need to do going forward. And you can iterate on this in a million ways. It's up to your imagination at that point.

This is all stuff for you. I'm obviously not an HR person, but Tony, before the next one, you can give me some more ideas here, and maybe I can work on some specific stuff so we can see truly how you'd do it. And maybe we can even try to do it live with the systems you actually have.

[48:14] Sean Williams: Tony, I might recommend that a lot of us have access to Copilot through the Commons or through the Okta portal, where I think we're limited as far as who has access to ChatGPT, right?

[48:29] Tony DeStefano: No, you're right, but I think whatever is the most powerful tool — I feel like ChatGPT has probably got some more oomph behind it. And the enterprise has been pretty open about giving you access to the enterprise account. You just have to name the business case for it, and it's a pretty low bar in terms of what they're asking for. So I think we could if need be.

[48:51] TJ Larkin: This real quick is when you think about structuring prompts. Again, you don't need to write prompts — AI writes prompts for you, if you even need a prompt. But this is a general strategy to keep in mind: role, context, task, format, and guardrails. Here's who you are. Here's the context of why it matters. Here's what you need to do. Here's how you need to format it. And here's what you should not do. That's a general framework. Honestly, I don't even worry too much about this anymore, because the AIs already know this and are pretty good at it if you ask it to prompt for you.

And this is a general idea of how I like to think about things: the humans should always be at the beginning and the end, but you want the middle to be AI. Have it do all the hard stuff. You do the human stuff on the front and the back, in almost anything — content or workflows, all the stuff you're doing.

[50:30] TJ Larkin: Let me go over real quick where this stuff is going. I don't think you're going to use Grokbot, but just to understand where AI is today and where things are going. If you haven't heard of Grokbot, that's Elon's Grok AI. It's all the rage right now, because it takes a lot of the stuff we've talked about today that is hard — and it legitimately is, not like you can't do it, it just takes a lot of effort and focus to get it done. However they did this, they said they did it in like three months.

What it is: it's kind of a version of Slack, where you can have different agents for different tasks. The point is it does things so easily. You talk to it like a person. Chief of staff is kind of the most common — you're my top advisor. And you just talk to it like a person, and it talks back to you like a person and takes actions like a person. It will just go do things for you. And I'm talking about complicated workflows.

For example, I like to make AI songs, which I didn't even show you, but making AI songs is pretty easy. And I listen to my own music for real. So now I come in here and I just go, hey, make me a song about this. I pre-wrote these ones about your company. I'll send it to you after.

So it's literally writing the song. It is then going into Suno manually — it uses its own computer, it has its own computer. It goes into Suno, which is the AI music maker. It puts all the information in there. It then creates the song. Then I've even done this where it will take the song and make a thumbnail and put the song on YouTube.

This used to take me half an hour minimum, and it wasn't good. My half an hour of work would be bad. Now it did fifty of them for me in an hour. They were amazing. It's unbelievable that you can just drop these things to this guy who just talks to you like a person.

This is where all of this is going: assistants that act like human assistants, that you can just talk to like a person, and it goes, oh okay, yeah, I'll do that for you. Oh, what do you think about this?

And then you can have different team members for different things. You have a writer, you have a growth person, you have an engineer, but you just talk to them all like people. And even crazier, I have another system — this is all personal, I actually have a team version of this, way more complicated — where you can watch the agents talk to each other in Slack. I literally did this this morning. I finally got it working this morning. A writer agent wrote an article and then tagged the editor. And the editor said, okay, I'll edit your article. And then he came back and said, no, I don't like your article, and tagged her back and said, you need to change these things. And then she said, okay, I'll change them. That exists right now. I did it this morning.

This came out in May, not even two months ago. Also, if you want to try stuff like this, Grok costs, I think it's down to twenty or thirty bucks a month. If you just want this personally — Tony, hopefully you think people should do this on their own stuff, it'll get you more psyched to do it in business. And if you morally don't want to do that, that's fine: Meta, the platform that a lot of people hate, just came out with a free version of this, like last week or two weeks ago, called Muse. So you can look up Muse AI, M-U-S-E, and it's very similar and it's free. You can tell it, go do this for me, go figure that out. And it'll talk to you like a person and do it. And it's on your phone.

Does anybody have any questions? Tony, how else can I help?

[54:29] Tony DeStefano: I think some people probably are like, holy smokes, this is bigger and more than I was thinking. But that's the point. It's like, hey, let's just jump into the deep end and we'll float around for a little bit. Think about this for a couple of weeks. Let's talk to each other about some of these topics. And then when we get together, we'll be able to dig a little bit deeper on some questions or some ideas after we've had some time to marinate on it. The point of this was not to make us all experts. It was to get us thinking differently about how some of these tools can help us do our jobs, and maybe some personal things, a little better.

[55:15] Tony DeStefano: TJ, I appreciate the time. It's a lot to cover. I gave you a very broad base to work from, and I think this has been a good introduction.

[55:21] TJ Larkin: By the way, this is what Grokbot did for me. It made all these thumbnails, it put all these songs on here. To me, this is just insane, that it did all this for me just by me telling it.

The key takeaway, for everybody: don't worry about it. I know I probably overwhelmed you with so much stuff. The goal there was to show you how exciting this can be. Start taking action. Start taking baby steps. Try to set up a project. Go drop a document in there. I would do the interview for yourself. Start doing the simple stuff. And then hopefully it gets you excited to do it, and you see where this can all go. That's the goal. And we can be much more practical, in person.

[56:02] Tony DeStefano: Well, great. TJ, I appreciate it.

[56:20] Tony DeStefano: All right, guys. Thanks, everybody.

— End of transcript —