Subscribe
Share
Share
Embed
Real business owners. Real bottlenecks. Fixed on camera.
Each week Paul sits down with a guest who's stuck — in manual work, disconnected tools, or processes their business has outgrown. They diagnose it live, map the fix, and start building. No theory, no fluff, no generic AI takes. Just the exact solution Paul would build for a paying client, handed over for free on camera.
Every episode is a case study your ICP can't stop watching — because it's their business on the screen.
Kwame Kwakye Asan (00:00):
Databases are not simple to manage. They're not easy.
Okana Graham (00:04):
If you have AI making certain decisions, how do you fix it?
Paul Kortman (00:08):
Is this the death of UI then?
Kwame Kwakye Asan (00:11):
I feel like everything shouldn't have AI.
Paul Kortman (00:14):
What are you talking about there?
Show ID (00:15):
Every business has inefficient processes, limited hours, and technology that really isn't pulling its weight. This is Connex the Dots, where real problems meet real solutions used by real businesses. So I'm Paul Kortman. Let's connect some dots.
Paul Kortman (00:42):
Today's topic, SaaS apps. They're adding AI and AI agents in their platforms everywhere, all of them. Should every app have an AI or agentic stuff in it? And where do you build your agents and why? And I think that there's going to be some disagreement here, both on where we build our agents, but also if apps should have AI natively in them. So with that, I'm excited to see what Kwame's going to bring to the table today. What experience do you guys have with this? Have you had clients ask? And I think where this really came up is this morning we were talking about a client that is using ClickUp and Kwame quickly mentioned, "Oh, by the way, ClickUp has AI agents." And I just kind of went, "Oh, another one." So just gut reaction, should all these SaaS apps, every SaaS app you use, whether it's your accounting software, your email software, your CRM, your whatever, your coding software, should all of them have AI in them?
(01:53):
Should they have agents in them? Kwame, go.
Kwame Kwakye Asan (01:57):
I don't think so. I don't know if it's the implementation or it's just having AI in every single app. It rubs me the wrong way. There's a lot I can say about DI implementation and how they put the AI to use inside the app, but I feel like everything shouldn't have AI.
Paul Kortman (02:17):
So that's in major disagreement with everything coming out from the Valley, from all the startups and SaaS apps and whatnot. Every day when I log into a new one of our stack that we're using and subscribe to, a new one now has AI enablement inside of it. And in fact, we're guilty of the fact that we've built an app ourselves and we have AI inside of it. So why not, Kwame? What would be wrong with having AI everywhere, whether it's just the implementation or just the core fundamental?
Kwame Kwakye Asan (02:58):
I think it depends on the use case. What is the AI doing inside the app? What can the next person use it for?
(03:07):
There's one app that I can think of right now that a client tried to show me that, oh, we have AI in the app now. And my question is, what are you going to use it for? And they couldn't tell me what they were going to use it for because the app is so basic that it does not need AI. Unless you just want AI to just write every word you want to say for you every single time and you don't touch the app at all, then you don't need AI in that basic app because the use case for AI is to help you make your work easier. This app is already easy to use. So what is the AI doing there? What is it making it even easier? You don't want to touch grass anymore? You
Paul Kortman (03:57):
Don't want to touch grass.
Kwame Kwakye Asan (03:59):
Basically, that's it.
Paul Kortman (03:59):
Okana, we haven't heard from you. What's your thoughts on this?
Okana Graham (04:03):
Well, maybe it's the Libra in me trying to see the balance in everything, but I guess it would depend on the specific software. And while Kwame was speaking, something did come to mind and it comes up a lot in my Notion sessions. One way I try to differentiate Notion and what that offers against other software is it's almost like a blank slate compared to some of the other software that's purpose-built. So if you put AI in that, it's like Kwame said, do you not want to touch grass anymore? Why are you really trying to automate? But with Notion, you have more fluidity. You can build exactly what you're trying to build. You have more control over it. So for me, I think it depends on the specific software, the specific SaaS app. I would say maybe offer the ability to integrate with an AI agent, but don't tell you, okay, we have this and it does this because we know we made a client that use the same software, but they use it in completely different ways in certain situations.
(05:11):
So it's almost as if they're making decisions for their customers before they know what exactly they want. So I would say give them the ability to interact with, choose their AI agent to do what they want, but don't give it to me. It's almost as though they're just dropping it in there to stay relevant because everything has AI right now.
Paul Kortman (05:29):
Should every database software, I'm going, like you mentioned Notion, we've got Airtable as a rough competitor there, and then we can throw in SQL, Postgres, MongoDB, Superbase. We could even get into Snowflake and your data lake stuff. Should your database software that are more open-ended, should they all have AI and agents in them?
Okana Graham (06:01):
I would limit it. So things like database integrity, looking into the health of your tables, that sort of thing, I would say have AI do that and make suggestions and then communicate to you rather than I'm always someone who likes the human in the loop element before you go out and have an agent be a cowboy making certain decisions. What about you, Komi?
Kwame Kwakye Asan (06:28):
I think this is where I would disagree with you. I think all databases should have AI. One, because databases are not simple to manage. They're not easy.
Paul Kortman (06:43):
What are you talking about there? A database is a database.
Kwame Kwakye Asan (06:47):
I mean, for us who are engineers who've worked with databases for a long time, if we pick an Airtable, let's say a client comes in with 100,000 entries in Airtable, we know how to filter things. We know how to look for something specific. We know how to sort and to find what he's looking for. But then the simple man who hasn't used Airtable for a long time will find difficulties, "Oh, I'm trying to look for this one specific needle in the haystack." That is where you need AI to help you with. AI can help you search for what you're looking for. It can bring it to you. It can give you a link to it and you can just open it.
Okana Graham (07:37):
What happens when it makes a decision? What's the fix like? How long does it take? What guardrails are going to be there? And for me as an engineer, I'm always thinking of when things go wrong, how easily can it be fixed? If you have AI making certain decisions, how do you fix it?
Paul Kortman (07:56):
Okay, so we're all kind of getting in the weeds here a little bit, and this is great. So yeah, we're jumping into this, but I want to throw a curve ball your way because we've got, okay, we just went through databases and that sort of thing. You guys have hinted at a simple app that's just there to do one function. I'm not sure what you're talking about, which one, but let's say, for example, I'm going to throw QuickBooks out there as a... It's an accounting software. I wouldn't call that simple, but accounting isn't simple. And I don't use QuickBooks, so I don't know what it is, but I know that it now has AI in it or available to it. And there's a difference here of if I were to use QuickBooks and connect it to my preferred AI chatbot, so I use Claude as my go-to, connecting QuickBooks to Claude, now I can ask Claude, do that search.
(08:57):
And so Kwame, coming back to what you were talking about of just I'm searching for something, I want to know, hey, how much money did we make last week? What was our top expense? That sort of thing. What seems out of place? And so data analysis totally makes sense, but I can do that with my chat, my local Claude, and then have it index that from QuickBooks. What's the benefit of having an AI native in QuickBooks that I can't do from an external Claude with access to QuickBooks? Is there any?
Kwame Kwakye Asan (09:40):
First of all, if you are using QuickBooks and let's say you are a one-man company, you don't need AI inside QuickBooks.
Paul Kortman (09:54):
Because you should know everything.
Kwame Kwakye Asan (09:56):
Because it should be easy for you to find what you
Paul Kortman (09:59):
Actually need
Kwame Kwakye Asan (10:00):
To find. Those who use QuickBooks extensively, big companies that use QuickBooks have accountants who know what they're doing inside QuickBooks, and these big companies will also have their own AI models outside of QuickBooks. So that would be easier for them, and it'll be easier for them to integrate their AI, QuickBooks, and other softwares that they are using. The accountants will know their way around QuickBooks. The upper level management can query their AI and get what they need, the information they need from QuickBooks. But if you're a one-month show or a small company, you don't need AI inside QuickBooks. So basically, if you look at both ends of it, I don't think AI inside QuickBooks is necessary.
Paul Kortman (10:50):
Okay, your thoughts?
Okana Graham (10:51):
Well, for one, QuickBooks was the exact application I was thinking of. If I was using QuickBooks, I wouldn't necessarily think of using the AI. I guess maybe to Kwame's point, if I was a one-man person. However, and the reason why I said it and I though of QuickBooks is there's a lot of persons who use QuickBooks and they're across different industries. So you giving certain offerings, how effective would they be able to already know what to offer with their AI rather than giving, let's say, an MCP where you can then connect your favorite chatbot and then you stare it out. So you use QuickBooks as a database and you use a chatbot to interface with your data because it's so wide.
Paul Kortman (11:41):
Am I to understand your take on this, Okana, is just give me MCP access and I'll translate that basically MCP, if you know anything about APIs, APIs are how software communicates with software without humans involved, without a graphical interface. A graphical interface is for the human, an API is for the machine, and an MCP is for the AI. So three different ways to interface with a software. And so QuickBooks, when you log into it, you're on your UI and that's how humans interface with it. API is how we do integration. So we're Zapier experts, and so Zapier integrates with QuickBooks and that does it over API. That's how that happens. And then the third area is MCP. And so when you connect Claude to QuickBooks, you're doing it over MCP. So when we throw out the term MCP there, we're basically talking about the protocol, the way an AI communicates with a software.
(12:43):
Is it your take that if every software had MCP, that would be fine? We don't need native AI anywhere?
Okana Graham (12:52):
Majority, yes, along that line. And again, just not to be a dead horse, but Kwame, your ClickUp client, if everyone use it the way it's designed or everyone use it the same way, it would be easy to do what you wanted. But I know from interacting with a lot of clients, 10 clients, they'll use the same X software, but they all use it slightly differently. I just think it's a nightmare to try to say, "Hey, we have this that can do that." And then nine times out of 10, that's how they come to us is, "Oh, we used AI, but it doesn't do this. How can we do that?" And that's when I do a search, does it have an MCP? Because if it does, we can do a Notion Agent, we can work with it in Zapier, we can use Claude, we can do different things.
(13:41):
So for me, yes, give me MCP, let me steer it down the path I want it to.
Paul Kortman (13:47):
Is this the death of UI then?
Okana Graham (13:51):
No, I don't think so.
Paul Kortman (13:54):
I like the deep breath and hesitation you both had of like...Don't want to go on record saying this, but no. Kwame, do you think UIs are done? No.
Kwame Kwakye Asan (14:06):
I don't want it to be, but because I think sometimes you need UI. And I feel like because of AI, we are becoming so separated
Paul Kortman (14:18):
From the grass
Kwame Kwakye Asan (14:19):
The grass…that we are not even noticing that we are getting separated from the graphs. Previously, if you wanted to do something inside Airtable, you had to visit Airtable. Now I don't have to. I can just type it in my claude. "Hey, can you bring up these two records in Airtable? "And I get it. So should we convert everything to MCP? No, I feel like we're getting to the point where we are just being distanced from everything.
Okana Graham (14:55):
Have an MCP to use to get to a particular goal, because I wouldn't see the sense of paying for a particular software, having MCP to please use any software. I still think there's some merit and some worth in UI because some persons, I feel like they just want to see certain things. Yes, you can get the information, but they might want to go back and verify that this dashboard says this. And Paul, we've had it where we've had conversations with Claude based on some internal dashboards and then we're like," Oh, let's double check this. Without an interface, you can't 100% confirm. So I still think it's not dead yet, maybe, because AI is how old? Four
Paul Kortman (15:48):
Or five years? For human use.
Okana Graham (15:50):
Well, I mean for
Paul Kortman (15:51):
Mass consumer
Okana Graham (15:52):
Two years. Consumer. Yeah. Yeah. So what is going to look like in 10 years?
Paul Kortman (15:59):
Oh, and now are we getting into a prediction show? I don't think any of us are qualified for that because five years ago I could not have predicted AI would be the topic on my parents' mouths. They're talking AI because, I hate to say it, but Instagram and Facebook have AI all over it now too, and all the content they consume is AI generated content for right or wrong or otherwise. So it's everywhere. I don't think we can predict where we're going to be in 10 years. All I know is that right now there's a huge trend that every app is putting AI in there. So there's a sub-question that I kind of hinted at the beginning of the show about agents. And so I'm going to take a stab and see if I can explain agents well. We started as a automation agency, so we built automations, and the basic flow of an automation is there's a trigger, something that starts an automation, and then there's a recipe that it follows.
(17:07):
And so when this happens, do this, this, this, this, this. If this, do that. If this, do this, then loop through that, et cetera. But basically you have something that triggers or starts it, and then you have a flow that it works through. We now call that a deterministic flow because every time that email comes in that matches that subject, it's going to run this flow and it's going to follow these steps in this order.
(17:35):
So we can put an AI step in that deterministic flow to determine, oh, should I go left or right? Is this a happy comment that the client is expressing or an upset comment? Is this a complaint or is this an urgent ticket that I need to create? Those are decisions to make inside of a deterministic workflow. So that's the next step up from a deterministic workflow is putting an AI question in there to be able to decide and make answers. Or for example, hey, write an email based on this input. We're going to use AI to make it more human sounding instead of just a raw template. So that's how deterministic workflows can use AI. But then you have the next level up, which is an agent where you set a trigger and you have a prompt, basically, a description of what you want the agent to do in natural language.
(18:36):
And so that's an agent, a trigger and a prompt, and then tools that it has access to. You can grant it tools. If anybody's familiar with how Notion does their custom agents, that is one of the most brilliant and simplistic ways for me to explain it to people because they have a trigger section, they have a what to do or a product section in the middle, and then they have an access section at the bottom. And so this agent is free to make up its own mind to do whatever it is based on the guardrails that you put in there. That's an agent. Now, that being said, apps like Notion have agents. ClickUp now has agents, which is new to me. I just found this out today. Claude has agents, ChatGPT has agents. Where should these agents live? Why do you have them live in certain areas, et cetera?
(19:38):
We all know on the deterministic side, you put that in Make, you put it in PipeDream, you put it in Zapier, Workato, these sort of iPaaS platforms. Where do you put your agents and where do you recommend clients put their agents?
Okana Graham (19:56):
For me, I think it depends on what your agent is going to do. And so it's two things for me. What's your agent going to do if they've come to us and they've had a particular stack already? So let's say they've been a previous Zapier client, then I'll try to build in Zapier. And specifically what it's trying to interface with? So nine times, well, majority of time I would say Notion because where it's trying to interface with, let's say, emails and specifying a trigger. However, if it's something that's behind, let's say an authentication screen or you're trying to gather raw data, I would switch to Mind Studio because you can do a web plugin and then I can activate by the click of a button and then I have my agent take action there. So it depends on the use case, on what content, what we're trying to achieve, if it's something gated behind authentication, and if we have a blank slate, I will go with the option that has the least friction.
Paul Kortman (21:03):
What about Make and Zapier? Because they both have agents now inside their platform. They've leaned heavily into this because they see the move away from deterministic into Agentik workflows. If you had to choose between those two platforms or a Notion or something else, what are the factors that would give you that choice? And/or Twami, do you agree with what Okana said is basically the least friction?
Kwame Kwakye Asan (21:34):
Yeah, I agree with what Okana said. Basically, depending on the stack that the client came with, that would inform where I build the agent for them. Within the past few months, I've grown to like N8N and the agent, just the presentation of the agent itself, I really like how you can choose which agent do you want to use, which LLM do you want to use, what tools do you want to give the agent and how they present it? Just how they present it looks the best to me between Zapier, Make and N8N, I would choose N8N. But depending on the client's stack, that would inform what I would go with. If they're already familiar with Zapier, I'll go with that.
(22:24):
If they're already familiar with Make, I'll go with that.
Paul Kortman (22:25):
When I was describing the Notion agents, you nodded when I said it just is a very simple three sections, that sort of thing. So I'm going to throw it back to you, NAN versus Notion on agent building, pros, cons.
Kwame Kwakye Asan (22:41):
So Notion is a simplistic agent.
Paul Kortman (22:45):
Is that a compliment? I would say
Kwame Kwakye Asan (22:47):
That agent in Notion is... Yeah, it's very simple. It's easy to use. Yeah, it's easy to use. Anybody can use it. When I have Notion sessions, one of the last things I focus on is a Notion agent because people just find it easy to use. I wouldn't show a client go and build an agent in N8N because those agents in Zapier make an N8N are for complex situations where you need an agent's input. But if it's a simple agent, oh, get this email, send this email, do this in Notion, I would build it in Notion. I don't know, what about you, Okana?
Okana Graham (23:33):
I mean, I quite like the interface with Notion, and I think that's a good thing that it's simple because in previous sessions, because of how simple it is, it's easy for them to make iterative changes. So I always say, "Hey, it might not perform the way you exactly expect it to, but here's where you go to make adjustments, and then you can speak with it in natural language and it makes the adjustment." I think some of the interfaces might be a little bit scary to non-technical persons. So again, it depends on the support case. Is it us that's supporting them? If it's us, then we have control over which one and there's no friction, N8N, make Zapier. And even Zapier, I would say, is a close neck and neck in terms of simplistic, the UI that they offer and the way you can make your adjustments.
(24:30):
So I would say Notion into that specific question. But one thing I did remember while you guys were speaking, another factor that determines which agent, I know one advantage of Mind Studio, depending on what you're trying to work with, you don't necessarily need an account with that software. So for instance, if you want to do graphics or audio, you want to generate that. On Mind Studio, you don't necessarily need to have an account with any of that AI agent. You just go through Mind Studio and then you can produce the same result on that AI platform and then ingest it in any other, whether in Mind Studio, and then you can also incorporate Zapier, Notion,
Paul Kortman (25:19):
Anything else.
(25:20):
I'll restate that for our audience just to make sure in case you're not aware. On Zapier, you can connect to your claude, you can connect to your OpenAI, and it uses your tokens, your cost, your subscription. You also on Zapier, you would connect to say your Miro account to create whiteboards there, et cetera. You would connect to your Google Drive account to create stuff there. What Okana is saying is in Mind Studio, you can do all of that, but you can also say you don't have a Miro account. I'm using that as just an example. You could use their Miro account to be able to create something to get that information back, or you don't have Claude, you can use their Claude account to be able to create something and get that back. Obviously, Mind Studio charges you directly for that instead of you having a separate subscription.
(26:17):
Similarly, that's available inside Notion as well that you can use different LLMs, but those are different models, but those are all similar competitors, so like OpenAI, Anthropic, Kimmy, Grok, that sort of thing. However, what I'm hearing from you, Okana, on Mind Studio is that you can use things that aren't just LLMs,
Okana Graham (26:44):
Specifically on assets, asset generation, so audio images. So on those platforms, you don't necessarily need to have... You wouldn't need an account, you just plug that module in and you say, "I want to use this on Google Veo or Banana, what's it called again, that one." But you just select that model and you can mix and match as you want there. The interface though, that would be a case where I'm building because it might be a little bit scary to the non-technical folks. However, after it's done, it's just a setup where you have a single account on Mind Studio and that agent achieves what you're trying to without you having to have a specific sign in to them.
Paul Kortman (27:33):
Great. So recommendations for clients, folks listening to this. Hey, we talked a little bit about, okay, AI is now in every SaaS platform. We talked a little bit about agents in all these different platforms. Where should clients start? What should they do? When you are working with a client, first session, they come and they're like, "I need this AI or I need this workflow or whatever." Where do you start? What's the first thing based on what we've been talking about today?
Okana Graham (28:08):
Well, for me, I would say I need to know your software, your software stack that you're working with, and an example of what you're trying to achieve. So I need to know a very good example of a good result. And then depending on if you come with a stack, like an iPad software, if you don't have one, then I make a decision on the least friction, the way I can get to that with the least friction, the easiest way to achieve that in as quick a time as possible. So I would say come with your stack and know what you're trying to achieve. An example, if it's a document or if it's to say in this scenario, have a predefined scenario and have it well articulated. Sometimes the ones that are difficult to work with is that they don't know what. They know they want to use AI, but they don't have a specific set list of things that they're trying to do.
(29:07):
If we have it set in the beginning, then we know what we're working with and what the end result would be.
Paul Kortman (29:13):
So have a goal, have an end result that you're trying, not just generic, I want to use AI better.
Okana Graham (29:21):
Yeah.
Paul Kortman (29:21):
Awesome. I was thinking also start with the platforms that you'd have, like you said, the software stack. And so what platforms do you have? What's available there? So when it comes to agentic work, what's available on the platforms you have? And then when you run into roadblocks, reach out. That's where we can advise how to get past those roadblocks, how to go from there. I happen to know a lot of users use Copilot inside Zapier to build their workflows, and then they run into a roadblock and they reach out to us. That's what we're here for.
(29:58):
I'll just leave that with that. That's one question down. If you've got any of your own, reach out to me on LinkedIn or to one of the guys on LinkedIn. Who knows? We may turn it into an episode. Subscribe so you don't miss the next one, and let's connect some dots.
(30:22):
So if you're listening to this and you want to understand more about how to practically improve the efficiencies inside your business, check out our other episodes. That's what they're here for. But if you're ready to start, reach out. We offer a no-cost consultation and I'd be happy to help you.