Claude Code, MCP, and the Future of Enterprise AI w/ Tonya Berenson

Speaker A: And my point is everyone is going to become a programmer, like, who wants to go into a UI and click around on boxes and try to find out where they put this on this click down menu, right? You just ask the chatbot. So what I see happening is the end of documents, the end of spreadsheets that's going to happen in the next couple of years. Those documents, eventually, as you work with more and more of them, those will get transformed into a vector database. 1, 2, 3, 4. You're listening to Owner RX with Alan Pence, the podcast for business owners, uh, who want to scale without sacrifice. Each episode delivers battle tested strategies from the Owner RX Playbook library, showing you exactly how to build systems that run without you. Here's your host, Alan Pence. All right, we are back on the Owner RX podcast, uh, with Tanya Berenson live from San Cugat, Spain. How are things there today?

Speaker B: Good, everything's good. Getting a little, getting a little chilly even here in beautiful San Diego. Yeah, I had to put the heat on.

Speaker A: Oh my gosh. We were actually warm over the weekend, so now, but it cooled way down today.

Speaker B: So that's why you're wearing the jacket.

Speaker A: That's right. I'm wearing my seasonal jacket. If you go to the YouTube, we decided it was the wrong color, but, uh, very good. So what are we talking about today, Tanya?

Speaker B: We were going to talk about AI agents and how do they actually play out like Claude code? Uh, for example, I want to hear.

Speaker A: We're running this. Tanya is talking about, we're running this system nrx where we basically use this coding agent called Claude code, which is really just the same thing. If you go to Claude AI. Ah, and use talk to it through the chat interface. Um, it's the same model. We're just using a different user interface which is either. It's like a terminal. So people get scared by uh, that, um, because it looks like computer programming, which I guess it is today, but it's really just like you could open up the terminal on your computer. So if you're on a Mac or I think it's the same on a. I haven't used a Windows computer in 15 years. Uh, forgive me, but, uh, you know, on a Mac you just go to your apps, you know, your uh, launch or I guess it's called apps still and just look for terminal. And I'll bring up this like command. It's called a command line interface with like the little cursor that doesn't blink. And everyone's scared of that.

Speaker B: And they're like, oh, I'm not going to lie. The first time, you know, when I saw it the first time, like there's just, I can't know. Yeah.

Speaker A: Uh, and then you can like, so you, then you can use it uh, in a different way. It's slightly easier. Is a program called VS Code, Visual Studio Code. And you can download that and that gives you an interface to it. You open up a terminal there or you can actually use that on your computer. And then you back everything up that you're doing onto what's called GitHub, which is owned by Microsoft. But it's the way that coders, they use that as their cloud storage mostly instead of Google Drive and uh, OneDrive. But GitHub also has like when you're in the web interface, you can actually open a code space which is just like another terminal. It's actually more secure to use that I think because, um, it doesn't touch your machine. Like when you're running the terminal on your machine, you're working on files on your machine. So if somebody gets in there or whatever, they could use that claude code to do stuff like steal stuff on your machine and things like that. So if you're in GitHub, it would only have access to the files in the cloud storage on GitHub. Um, so anyway, that's just a long way of saying what I see happening is over time, like my line has been, uh, people worried about programmers today. Like, oh, all the programmers are going to lose their jobs. And my point is no, everyone is going to become a programmer. Everyone, basically everyone who does any kind of work that involves words or symbols and software is going to become a programmer. And that's what. So you are now Tanya, you are a programmer once you use that system.

Speaker B: Can you believe that?

Speaker A: Uh, Tanya can do it. You can do it.

Speaker B: I'm a programmer. Baby steps. But yeah, I mean even using Visual Studio you do have an extra step to have to get everything that you program over to GitHub. But I've learned to do that as well.

Speaker A: Yeah, so you push it to GitHub. You've committed to GitHub. See, you're a coder now.

Speaker B: Yeah, I've committed to the coding process.

Speaker A: Yeah. Now that part of it I think is going to go away. But what I think is happening is that ah, if you think about it really conceptually, all programming ever was was a way to communicate with computers, humans to communicate with computers. And it's gotten easier and easier from early kinds of C and I don't know. I don't know the history of computer science, but got easier and easier, like Python and R and things like that. They're just easier versions of ways to create programs that help humans talk to computers. So now we're going to a point where it's gotten so easy with AI that you can just talk to it. You don't need someone in between creating a program to translate. It's sort of like, I don't know, up till now you've needed a translator to talk to someone in France if you don't speak French. But now you can put on your Apple, uh, AirPods and it can translate in real time for you. That's what's happening, right? Essentially what they have. Yeah. So you get to talk directly to the computer. There's no priestly cast in between who know the secret code and have to translate for you. So the reality is so like, to me, that doesn't mean that coders go away, it means we become coders. Right. So what you're essentially doing in Owner rx, we have workflows that I've built with cloud code. And I just tell it like, here's how I want to handle the meeting notes from our weekly meeting and I want an agent to extract all the summary and the action items. I want another agent to look at like other meetings that happened before and make sure we haven't missed something over three or four sessions. I might want an agent that compiles all the meetings into one big summary. So if you want to do like a quarterly meeting, it would do an analysis of the whole quarter you got. I mean you can do whatever you want with it. So yeah, you know, I have that. I have like marketing agents that, that have rules about how to create LinkedIn and newsletter posts and ask me questions of things are unclear. I have um, one that like when we do research on something, so we're doing research right now on how to build better newsletters. You put the content in that we're reading or looking at and it does like a copyright analysis. It uh, extracts the key ideas, make sure that they're generally known ideas. So I'm not like stealing anyone's ip. If it is someone's ip, then it notes it and says, hey, you can only use this much of it under fair use. Um, does all that for me. Um, and I have, you know, we have like probably five others right now. Uh, we're running a pretty simple operation pre launch of the software. So like we're not running dozens and Hundreds of agents yet. We will probably. Um, so the question I had really over time was, since everyone's in there just talking to the computer, and the computer has access to the data, which is right now a bunch of files and GitHub. So this is one thing that's. Yeah. So over time that, you know, that would become too big for Claude code to really understand. It couldn't digest. Like, even now, my. I like the research agent and my coaching agent and, um, the marketing agent, all basically provide the documents at the end. And then there's another agent called the Pipeline Agent that comes through once a week, looks at all the things that were done in those three places and extracts new insights to put into my playbooks that go into the software.

Speaker B: And you created the Pipeline Agent as well. You've created all of these.

Speaker A: I just told Claude, create an agent that does this, this and this. I tested it out, it didn't work right. So I told it, oh, do this right. That's all I'm doing. I'm natural language and it's building it. Right.

Speaker B: Yeah.

Speaker A: Um, there's some tips you need to learn. Like, hey, make sure you plan really well up front. You know, they're basic, like, project management tips. They're not programming tips, but anyway, so I built that. So that repo, though, of all the content of my coaching and my insights and what we're researching, that's starting to get really big right now. It's just a bunch of files, like markdown files, which are like word files that are organized very well. Um, so like, what's going to happen in the future is those are going to have to go into a database. And that's scary to people, too. They don't know how to do that. But. But essentially there's like, vector databases, which are what, uh, large language models. AIs are really good at reading. So it relates. So it would take like a markdown file that has like five subjects with like a paragraph describing whatever that concept is, and it would break it into five separate pieces that are linked together through the database, live separately. And then when the AI comes in, it can bring out all the different pieces. They're vector related to each other. I don't know exactly how that works.

Speaker B: Is this like a step, Allen, when, like, Claude code runs out of space? Basically, yeah.

Speaker A: Well, Claude code can't, like, run that many files, right? Yeah, it's too big a set of knowledge for it to, like, just run over real quick. It needs, like, it needs structuring to, like, know where to look. So this is like a vector store is like it's concepts instead of numbers from your transactions, from your point of sale software. So that's going to go in a regular database. But how does ah, AI agent relate to project management? That's going to go in a vector store? Because those are two different concepts. Um, so that just basically what's going to happen is everyone's. So what I see happening is the end of documents, the end of spreadsheets. That's going to happen in the next couple years. Now that we're doing this. I haven't even gone to our Google Drive in months. I don't go there. Why would I? There's nothing there. All the documents are in GitHub now, those documents, eventually as you work with more and more of them, those will get transformed into, into a database, a vector or regular database. Right. And then you're just running the agent, you're running your cloud code in a chatbot. Basically a terminal is just chatting with a bot. Right. So I think you're going to go to not what we have today. Vs Code GitHub Codespaces terminal. There'll be some kind of web interface that has a chatbot and has like probably a dashboard on the side and that uh, dashboard or you know, panel there will display because like right now like a chatbot can only display so much, right? It's telling you stuff. So it's not a great. Like you don't want a dashboard coming out of ChatGPT, right? You want to see it on the side. And you see this ChatGPT has been doing this. As you search it's starting to create things on the side and leave the chat there. Right. So think about it. Why do you need. What are documents and spreadsheets? They're like where right now they're where you do work. So like you're typing into a document or working on a spreadsheet. Okay, well very shortly right now you can go into at least Gemini in Google Drive on your sheets or your docs and tell it what to do and it puts it into the doc. So why do I need the doc? Right, to do the work. The other thing they are, are representations of the artifact that you're looking for. So like the paper you want to present, the slideshow you want to present, the Excel spreadsheet you want to show.

Speaker B: Yeah.

Speaker A: The visualization of it. Well the reality is these coding, these coding agents like CLAUDE code are increasingly getting things called skills that allow them. They're basically programs. So Claude code will have a spreadsheet document, any kind of program skill, right? It'll have that and it'll just spin up whatever you need at the time. So why would it take. Why would you use documents that are sitting static in a drive, cloud drive somewhere? You don't know what version they are, you don't know what a blah, blah, blah. You can just access the database, the vector or the regular, and it knows the latest status. And then the document or spreadsheet just shows up on your pane and you should be able to send it to people and stuff. But it would update in real time, right? So it's just a document is just an artifact of some kind of process that can live in a database. And so as that as AI becomes easy for humans to talk to the computer and ask it questions, why do you need these fixed things that are substandard for the job that you have? So my take is that all goes away and you're just talking to the chatbot, uh, with a pane next to it that shows you where you would

Speaker B: want that to go away. I mean, I'm sure there are people out there who enjoy entering numbers into a spreadsheet. I'm certainly not one of them, so. Or any kind of data, you know.

Speaker A: Right. And then they're building all these connectors to other apps, right? So these like, um, Model Context Protocol MCP is the new standard. It's like an API application programming interface. It's a way to get different computer programs to talk to each other. It's very, you know, it's pretty amateur right now. Anthropic came up with this. The guys that make Claude, everyone's adopting it, right? So pretty soon that chatbot, you're already seeing it, right? You go into connectors and ChatGPT now, and it's all NCP connectors to your Google Drive, to your Gmail, to HubSpot, to whatever you use, right. And so QuickBooks. And so, you know, you're gonna have this chatbot, uh, that has access to the databases of all the information that you use and your company uses all kind of like permissioned to like, only give the information to the right person. That's going to be another step we need to go through. Um, but, uh, like I just see the end of files. Like you're just going to be working in the chatbot with a representation on the side. Like, who wants to go into a UI and click around on people's stupid boxes and try to find out where they put this on this. Click down Menu, right? You just ask the, you just ask the chatbot. Right? Um, uh, and I think all these people who are saying, like right now, the line I hear a lot is bring the AI to where the workflows already exist. So that means, hey, I'm in Slack, I want to mention an agent and have it do work. I'm in Google Docs and I have like a button to write my newsletter. I guess that just seemed to me that's like, hey, uh, let's upload, um, our brochure as our website in 1997. It's like the same level of sophistication and like BO, but we could have web apps that are interactive and we can actually sell stuff on our web. That's what was coming, right? And I think right now we're still trapped in this paradigm where it's like, oh, we're going to bring AI to these substandard, um, interfaces that people don't really like anyway, instead of just jumping over that.

Speaker B: Jumping over that, right?

Speaker A: Yeah. Like, you're not going to log into HubSpot. You're not going to log. I mean, you will in the background, right? You won't even, probably even know about it at some point. You just log into your front end and I'll be connected to all these things and you'll just ask for stuff. And like I, to me, anyone who's building something that doesn't do that, I mean, I guess there's a period of transition. Right? Of course. So, you know, but if you're not looking at that long term, um, you know, I think that's crazy. Like that's where we're going, clearly.

Speaker B: And it's not that. I mean, it's already happening.

Speaker A: I mean, we're doing that, right? We're still using Documents right now, but pretty soon we won't. And I'm about to transfer, you know, like when, uh, when we go to the tool, all my knowledge base is in a vector database using through Owner rx. And so basically Owner RX is going to become that thing that we like. We would use it, we'll use it to run our company.

Speaker B: Right. No longer will be K. Allan, can you send me that document where you outlined X, Y and Z? Right.

Speaker A: Where is that document? What is it? Or where is that information?

Speaker B: That's, it's more that like, where's that information?

Speaker A: And then, um, you know, we still have some stuff we got to do where you got to make, you know, there's some, uh, a few hurdles to get over to make this easier to do. And like right now, this system we're running with like uh, a terminal, a command line interface, put it that way. That's what they call it, which is what the terminal is. And um, GitHub is not. It's kind of janky, right? Like GitHub is not fun to use. It's like the setting up all the permissions and um, issues and pushing uh, and like these um, pull requests, PRs, all that. It's just like another language. It's another language that no one wants to learn. And it works well if you set it up right. And we're actually about to do it. We're launching a commercial service for AI and automation, customize. Whereas Owner RX is sort of like a SaaS solution. Um, but we're doing that for bigger companies. At my um, former company Corner alliance that I still own but don't run, we're helping set that up. And um, we'll probably use like this GitHub, um, uh, Claude code sort of set up for now. Um, until like the tools really get to the place where you're just working in the chatbot with like the pain next to you and hopefully Owner RX gets there. Uh, um, but there are a lot of people working towards this. And really for Owner rx, we don't really care about where the agents are running. You know, that's what everyone's going to fight over because that's the infrastructure.

Speaker B: What do you mean, where the agents are running?

Speaker A: Yeah, so like Google, Microsoft, OpenAI, they all want you to build the agents and their kits like that they provide and then that means you're going to run it on their servers. So you're going to use their cloud storage. Amazon will want that too. Right? And that's why they're all doing these deals behind the scenes. Where are the agents going to run? We don't really care. Right now we're running them. We're actually using aws, but right now we'll run them because there isn't a great solution that works for everybody yet. Uh, there will be at some point, but at that point we'll give up running them. We just want to be the interface that helps you figure out what agents to use and then consume the like, overview of what the agents are doing. So like we want to be the fate. We want to be that like chatbot, uh, pain. We don't want to be the background like running the agent. We want to be the place. And then our agents, the way we set them up are based on my knowledge and my continuing research and work on it. Right. So we become this very small business specialized agent factory where you select them and use them. Um, but like as far as the infrastructure of them, I don't really care if they run on Google Vertex or Microsoft Copilot or Open AI or whatever Anthropic is going to have. Right. I have a feeling that the way these guys are going right now is this sort of like midterm to long term play that's kind of difficult. So in the short term, the medium term you see like Microsoft wanting to run all the agents out of their existing stack of applications. So Word, Excel, SharePoint, OneDrive. And I don't see that as being optimal for the future. So maybe that works for the next year or two but then everyone's going to move to just this code interface, some kind of chatbot interface that you talk to it and get stuff done and then have the pain come up and you can share the pain with other people. And that's, I mean a pain. P A N E not P A I N. Um, so I think that they're trying to play this game of like what would people adopt now versus long term. And I think like Anthropic is really well set up. Claude is really well set up for the long term because of the way they're going with skills which are these little executable programs, small programs. Right. Um, but Microsoft is really set up for today. So it's like sort of this, which one is going to win out when. Right. I think I know what the long term is. I just can't predict like how long we're going to be in this medium place where like you know, if you get. Because like we're already there sort of with maybe like the vector database not being in place yet. But um, others aren't going to adopt what we're doing. So they'll probably be in that uh, Microsoft Google system. Google's sort of in between. I feel like they're kind of doing both. Um, but yeah, I think like long term, if I were really gearing up as of now, I would say go learn how to use Anthropic. They seem to be the ones, Claude. It seems to be the ones driving toward like an enterprise company like B2B kind of way. Whereas I think OpenAI is trying to do everything right. But they're really a consumer chatbot search kind of. They're going to be an advertising driven consumer product. Microsoft is in a weird spot. I don't know like yeah.

Speaker B: What? That. I can't really get my head around that. Like what?

Speaker A: Well, I guess my question is, do Microsoft and Google become just like hosting environments? Like they're really just, um, they have this suite of productivity software that I feel like is going to get commoditized. Word is just a skill that Claude can run. Why do I need a separate program called Word that feels to me like that gets commoditized pretty quick, like CRMs. I don't know why I need a CRM. I can just have a database, right? That's all CRM. A CRM is a database with a user interface.

Speaker B: It's the need for it, right?

Speaker A: Why do I need that? Why do I need this stupid ui, right? If I could just port all the data out into a database, why am I paying Salesforce a bunch of money? And Salesforce thinks it's going to run all the agents out of Salesforce. I disagree. I think they are just not going to have a model like Claude. Now maybe they get Claude to come in there, but eventually, eventually it's going to be chat interface with CLAUDE to some kind of database set of databases. That's it. And with a pane that shows you stuff, I think everything else gets wiped out over time. And like even like complicated ERP type stuff just becomes a series of skills.

Speaker B: What's erp?

Speaker A: Uh, enterprise Resource Planning. So that's like, you know, like at a bigger company, like a NetSuite or something where they have like the general ledger that does all the accounting. They have like a purchasing system, a budgeting system, you know those, those will just become skills. I mean the software is so easy. Software is getting easier and easier to create and like, why are you going to pay extra for something that's just like downloadable to Claude? Um, so I see it going to that. Yeah, yeah. So maybe, maybe you have Anthropic up front. But then like Microsoft and Google are providing all the compute, right. All the runtime. Like Claude's actually the language executing. But they might. But they're using right now they're renting like either Microsoft, Amazon or Google servers. They're not necessarily building all that themselves. Like OpenAI is building some of it themselves. That's why you hear about all this stuff. But Anthropic doesn't have that kind of money. They're not doing that. Um, so they're going to partner. So to me, like Microsoft probably becomes a commodity provider of compute to somebody like Anthropic. But that's more complicated, takes a little more prognostication than I know. I just know that this is where we're going to is like chat interface or verbal. Right. You can just put a verbal interface into that too where you just talk to the computer. It does stuff and gives you visual representations of it when you need it.

Speaker B: I m mean it may be a little intimidating, but it's really, really cool.

Speaker A: Yeah, it's awesome. It's so much better. Who wants to click around in someone's stupid UI? Uh, like some 25 year old from Silicon Valley decided this is how you have to access your information. Screw them. You can do it the way you want.

Speaker B: Yeah, screw them.

Speaker A: That's right. Um, yeah, I think this is where we're going. I think it's really important if you want to get there faster. I think I might put a little guide up on how to use GitHub this way. I think it is very difficult. Like I did be helpful. I did a LinkedIn post on this and it went a little viral and I got a ton of comments about it. Some like, oh well you could just do it. And some like, oh yeah, well this doesn't work very well. And it's like. And some people are like well you can do it. And I'm like yeah, the point isn't that I can. My point was like it's really easy for me to use the system and I can train like one or two people to use it. And it's not overwhelming but like if you're trying to scale this to a 50 person company, it's tough. People don't like it. Yeah, you got to get really good at setting all the permissions and everything. People do not like it. It's not familiar. It's just like right now I wouldn't recommend it for an enterprise but if you're building something yourself or you're starting up, you have a new company or you're kind of working on things that you don't necessarily need the entire company to interact with. I think this is the most get

Speaker B: in from the ground up.

Speaker A: Yeah, you can just build agents by telling agent all agent is a series of processes that I do, uh, repeatedly. Right. And so I just go in and tell it, okay, do this. And so the advantage is now with my newsletter, say LinkedIn or other agents I built in like antagonist agents that check the work of the first agent and it's really effective and multiple times I've everything's gotten way better because I had that built in. And so like that's what you can build in now. And you can start running your company with, like, half the admin staff that you have. Um, and then as I can connect into more and more software, as these MCPs become easier and more available, um, you know, it just. The more I can do more, it gets easier. It just gets easier. And like, the other day I was like, oh, connect to QuickBooks. And it just built an API. Claude Code just built the API.

Speaker B: And you literally just said, I want to connect to QuickBooks, build me an agent.

Speaker A: But how long until it's just a skill, right? Or just a downloadable ncp. Or it just comes with Claude, Right? They have a connection. Right? So that's coming within months. So I think we're just going to get there sooner. So I urge people to experiment with it, but I'm not sitting here saying, hey, you guys should all go out and implement this in your company. I just think it's a bridge too far for people right now. But if you want to see what the future is going to be and start getting there yourself, I definitely say experiment with it. Right? And that's what I would do, is use Claude code and my GitHub code space, the most secure way, set up a GitHub account. Anyone can do it. And then have Claude teach you how to use GitHub. So give me five lessons on using GitHub better and it'll teach you.

Speaker B: Yeah, anybody could do that. Really?

Speaker A: Yeah. And it's intimidating. I know you look at it and you're like, oh, it's so confusing. But, like, just take a screenshot, put it in the ChatGPT and say, where do I start? And it'll tell you. So, um, yeah, so I think that's it for this week.

Speaker B: That's awesome. This was helpful, useful.

Speaker A: Cool. All right, M. Until next week.

Speaker B: Adios.

Speaker A: You've been listening to Owner RX with Alan Pence. Want to apply what you just Learned? Try our AI Business Advisor at OwnerRx. It has all the insights and lessons from this podcast, plus hundreds of Playbooks. Ready to solve your specific business challenges. Owner rx, stop being the bottleneck. Start being the owner. Want to hear you go.