Agency Forward

Chris DuBois (02:25.186)
How should agencies be thinking about AI?

Miles Ukaoma (02:44.249)
Yeah. So, you know, some, some of how we discussed this prior to getting started here was if like AI should be seen as a threat or an opportunity. And it's, it's becoming played out this trope I'm about to say, but we've all probably heard that AI is not going to take your job, but somebody that's utilizing AI might.

And I kind of view the opportunity with AI as a, an agency standpoint, but as well as like, if you're working within a brand very similarly, which is that like, these things are all multipliers. They're going to help us unlock a lot of productivity and hopefully like in an ideal world, allow us to do work that we feel find to be more fulfilling and more impactful organization. Like there's lots of things that we're doing on a day to day that

are pretty repetitive, they're pretty mundane. They're not driving tons of value for us or the organizations that we're working with. So the hope for AI and why I think that there's a massive opportunity that's there is that there's so much productivity that can come at the heels of adopting these tools correctly. And in an ideal world, it's also kind of elevating the quality of the work and the types of work that we're doing in that it's more focused on delivering value versus just like covering...

all of the table stakes or pushing paperwork or that like. So I look at it as a major opportunity. And I think that the larger threat is for agencies that are not adopting AI or not thinking about what their AI play is in the next five, 10 years.

Chris DuBois (04:19.074)
Yeah. we've talked about this enough times. Like I know we were going to agree on, this stuff. The, the threat side, it's like one of the things that gets me the most, like to the point where, I wrote, I think my last newsletter was about this. I see so many agencies that rather than seeing it as an opportunity to increase productivity, they see it as a way to cut costs. And so they,

Miles Ukaoma (04:45.349)
Right.

Chris DuBois (04:46.018)
They shrink their team down because they say, Hey, that one person can do the work of four. So I only need, you know, a fewer people, but then the output is the same as what their team was before. Right. Like, so you actually didn't like you reduced the potential multiplier here. And so the, the total output is actually a smaller number. Why don't you keep the same team and now quadruple your output and potentially have even better. Right. And there's also like, if I have three people thinking over an idea,

Miles Ukaoma (05:02.288)
Right.

Chris DuBois (05:15.616)
and being using AI to like do whatever, but they still have to do their regular tasks and they have to do all these other things. They're not going to be as like unique in their thinking and their ideas. Whereas if I had 10 people who could share that workload while thinking, Hey, how do we actually like make this better? How do we optimize, find new ways to do things? Like you're going to get some better results. And now your agency has like a distinct advantage over everybody else.

Miles Ukaoma (05:39.417)
It's like what you're saying is so interesting to me because like I see that all the time as well. Like where the shining beacon is actually kind of like dystopian, like in the AI space, like people are like pointing to like, they're like, Klarna, Klarna is like the wet dream for all of the companies. Like they want to move to what Klarna is doing. But when you actually look at like the base case of what Klarna has done, they cut a third of their support staff and replaced it with these, these AI agents.

I think that like it's created some misconception of like how these tools should be deployed. Right. Like I think lots of people look at AI from a gain of productivity in the sense of like, yeah, I would just be able to, you know, reduce my, my, my head count that's here. But like the secondary piece that you were touching on around what makes the output unique, I think like needs to be double clicked on because like these LLMs are literally created.

to predict the most probable word that's next inside of the sequence. So if you just think about like what's going to happen naturally, and they've been trained broadly on the internet. So like, if think about what's gonna happen next, we're all going to march towards the singularity, where all of the ideas start to look the same, all of the design starts to look the same, all of the content starts to look the same, because there's no differentiation. And like, that's where the people come in. It's our...

proprietary data that we have. It's our unique frameworks. It's our philosophy on how we view the world that actually allows for outputs from one agency to be differentiated from outputs from another. And like, if we're relying on the tool in this way of like, well, you just use this to replace all of the busy work. doesn't matter about the unique insights that you're putting in. You're going to get very generic outputs that

don't differentiate you from your peers. So there's definitely a right way to do AI and there's so much to press into with the wrong ways and some of the misconceptions that are there as well. But I am curious as to what you're running into with others that you're talking to, if they view it more like Klarna or they view it more of like, no, we're gonna use this to further differentiate.

Chris DuBois (07:47.362)
Yeah, most are trying to cut costs and speed up what they're currently doing, but they're not thinking about it from a, how do I maximize like efficiency in order to like increase my potential innovation and stuff. They're, trying to like use AI to basically replace themselves in the creative aspects of what they do. And so they're actually reducing their, you know, the potential, the reason that people are going to them and hiring them.

Right. The same reason you wouldn't, you don't outsource your main thing, right? Because that's why people are hiring you. You outsource all of the stuff you don't want to deal with, right? You wouldn't hire a VA to go directly to your like biggest clients and stuff and just run those relationships, but you would probably have them run your emails, right? Like set up your work schedule, like do all those other things so you can focus on the important tasks. But I think a lot of agencies specifically are like, well, how can I do design faster? How can I build this website faster?

Miles Ukaoma (08:24.71)
Right.

Miles Ukaoma (08:38.373)
That's right.

Chris DuBois (08:46.318)
It's like, what if you just use AI to set up the project for you so you don't have to spend an hour, two hours, making sure everyone has the right tasks in front of them and they can just sit down and start working and actually doing the things you hired them for. Yeah, that's what I'm seeing a lot of. I don't know if that resonates with your side.

Miles Ukaoma (09:03.833)
Yeah, I think that like more or less it definitely resonates that on my end. It's a shame though, because like that's not where the opportunity is in my mind. If we were to think about some of the skillsets that an agency would need to adopt in order to take advantage of this, we just got done at South by Southwest last week and go figure.

almost all of the talks were geared towards AI. Like everyone was talking about AI. But we went to one talk that I found to be like actually really enlightening in which the speaker was kind of bringing up this idea of skill relevance. And like I never had really thought about AI and its impact on what it is that we're doing on a day-to-day basis until like I'd seen it in this framework of skill relevance.

And in times past, like our parents generation, they would have like a single skill that they would refine over their lifetime. And then that would maintain their entire working career. So it's like, I'm a paralegal. And I was a paralegal for 40 years. Like I learned these base skills around typing, organization and the like. And I was able to monetize that skill for my entire career. Like fast forward into our generation, it's like 10 years.

Like you can gain a skill and you're able to monetize that skill for like 10, five years. And what the speaker was kind of touching on was AI is cutting that skill relevancy in half, like every 12 months. So we're to end up in this world in which the skills that you've developed as an agency can really only be monetized for a small window of time.

So if we're in a landscape that looks like that, what's the most important skill that you should be working on? Like as an agency, it's not some nuanced technical skill. It's actually the ability to learn and adapt and to change with, with the wave of, of, of the market and the wave of AI. So like, I, I, I'm kind of curious as like what you think about. Like this idea of like skill relevancy and like how AI is impacting the different capabilities that we think about as an agency that we want to monetize, go out and, sell services on.

Chris DuBois (11:23.086)
So it's pretty intimidating, right? For anyone in the agency space, especially those who are super reliant on tactical deliverables, like you will pay me to do this. I'm just going to keep churning out these assets for you. think the agencies that have found their niche in strategy are the ones who will that skill relevancy is probably not going to shrink as much per for every like iteration of this. and so more credence to the idea of like,

Miles Ukaoma (11:41.553)
Mm.

Miles Ukaoma (11:46.96)
That's true.

Chris DuBois (11:52.674)
get really good at strategy so that even if AI can do all of these other things really well, really fast, you're the one who knows how to put them all together. And eventually AI will probably be able to do that too. But for now, having that distinct like industry knowledge, you your niche knowledge of understanding what problem set you're working towards and being able to like apply all of these different things at once is probably what's going to save agencies or the agencies that do it.

Miles Ukaoma (12:19.279)
Yeah, think so too. For the agencies that are willing to do it. You know, it's pretty big shift, especially for those agencies that have been kind of set up to be very tactical. It's like we do website migrations. Like you come to us with a website that's on a different platform and we migrate it into the platform that you want to work on.

Chris DuBois (12:26.03)
Mm-hmm.

Miles Ukaoma (12:40.541)
Those are the types of, like if that's what your bread and butter is built off of, those are the type of agencies that seem as if they are ripe for disruption in this next kind of round of AI innovation.

Chris DuBois (12:54.062)
But I think there's probably something you can add. So I take website migrations, right? If you're an agency that's doing this type of work, like, yes, there's a good chance that AI is very quickly going to replace what you're doing there. But if you've done hundreds of migrations, you have seen all the things that can go wrong and all the things that could actually make a site better with this migration process. And so now if you can leverage those tools to make the, to reduce costs for how you migrate, but then throw on top of it, the strategy of like,

Miles Ukaoma (13:10.331)
That's right.

Chris DuBois (13:23.916)
You don't just want to migrate. You actually want all of these things on top of your migration to make sure your site's actually in a better position. And like now you've, you know, put off that, that relevancy gap that's going to pop up by at least a little bit. And so that's keep you keep yourself relevant anyways.

Miles Ukaoma (13:40.687)
That's exactly right. Yeah, it's a really interesting way of like, just repositioning. And it's like, it's not necessarily as if like the outcome is much different, right? Like ultimately that

in that engagement of the migration, the client wants to be able to get onto the new platform and have something that is functioning there, right? But like if you're able to bring in that differentiation of like, well, on top of that, have you also thought about, you know, how this website aligns strategically with you guys as Northstar? And here's all the different things that we offer from a services standpoint that help to further that advancement of that Northstar. Like that's where your real opportunity is, is like what...

on top of what you are usually selling in, what can you upsell into these engagements that differentiates you from the base level output that an AI would give you, which is going to give you the same outcome. It's just like there needs to be something more that's there to differentiate you.

Chris DuBois (14:38.946)
Yeah. And I think that's where even honing in on that specific problem set you can solve for and just getting, cause like most people will just say, you just need a website migration. Okay. That's a problem. like, yeah, but why are they doing the website migration? Right? Like there's an underlying reason, like what's going on in their business causing that migration happen. So if you can also help them with the like other things going on within that, now you're giving yourself more opportunities to be, you know, be seen as valuable. So, so

Miles Ukaoma (15:05.777)
100%.

Chris DuBois (15:08.224)
Okay. We got into some of the mindset stuff, which was, really important for me to make sure everybody is thinking about challenges correctly. and I don't know if correctly, just thinking about it differently. I can't say that I always have the best ideas. but I also want to get into some of the tactical, like how can we actually be deploying AI? What are some of the tools? How, what are some of the, that, I don't know, workflows that you could be adding within your day to day to make life easier. And you're the perfect person for me to talk to about this.

So open forum, what do you got?

Miles Ukaoma (15:39.535)
Yeah. All right. So, you know, we have a center of excellence around AI in our team, and we like to kind of experiment with new tools. And this month, month of, I guess I shouldn't really say the individual month, but this month, the big focus has been around AI tools that are voice first. Right. So I have several tools that we've been kind of employing over the past couple of weeks that

Chris DuBois (15:55.042)
Yeah.

Miles Ukaoma (16:08.625)
help us get more out of LLMs from an AI first input. like, I have a really strong feeling that voice is gonna become like one of the leading inputs for interacting with AI. So thinking about like voice first interfaces are gonna come, it's gonna become like similar to if you're working in web from an agency standpoint, like back in like 2010 where everybody was talking about like mobile first. Like I think that voice first is going to become a really, really big.

a focal point of like where this technology is going. And the tools that are like top of mind for me, there's three, granola.

I really am like loving this tool, Granola, which allows you to kind of take all of the information that's coming across your sound card and transcribe it into notes. And you can collaborate with the notes in that in real time while we're having this conversation, I could be taking notes off on the side and then the transcript is just going to go back and enhance them or fill them in. like where that's saving me a ton of times is like.

Chris DuBois (16:46.872)
Thank

Miles Ukaoma (17:12.557)
Every day I'm just in calls from the start of the day until the very end of the day. And it's easy to lose the context of what was important, what was decided, what we were going to work on, what the action items are, who needs to be followed up with. And Granola gives me a very comprehensive view of all of those conversations that are taking place. You there's lots of tools that are out there like Fireflies and Fandom that do something similar, but they have to be invited into the platform.

So like when you're in an agency, when you're in a marketing leadership role, you're having all kinds of one-off conversations where somebody just gives you a Slack call. Somebody just picking up the phone and giving you a ring and you need a means to kind of tie all those conversations together. And I think Granola does a really good job of that. And then, go ahead.

Chris DuBois (17:58.21)
Yep. Well, so I'm using Ernolla right now on this call, so that I can lead and I've been using it after, man. Okay. You introduced me to this last week and I have, I have now used this in every client call that I've had this week. I have the Fathom recording that's natural, but like even in there, if I go back to see what did we talk about last week? It's like, I have to look through the notes and kind of figure it out versus opening up this Ernolla thing that has my notes that it just like builds upon.

Miles Ukaoma (18:25.508)
in.

Chris DuBois (18:26.188)
And it's like, it's way easier for me to just pick up with, with a client. And so now my, amount of time that I'm spending figuring out like, where are we? What did we, you what were the required outputs from each meeting? What do I need for inputs for the next one? It's just so much faster.

Miles Ukaoma (18:36.955)
like that.

Miles Ukaoma (18:41.777)
Totally, yeah, like totally. And I think that like that's what we're trying to do with AI. It's not like how can I completely replace myself so I don't have to work anymore. It's like, how can I?

reduce the things that I don't like doing so I can spend more time on the things I do like doing. Like I don't like sifting through all of my notes to figure out what my action items are. So granola fits in as a really strong tool in the toolkit of AI. The next one that has been like top of mind for me, Chris, is whisper flow. And I think that you were actually the one that told me about whisper flow. So we traded on granola versus whisper.

Chris DuBois (19:12.856)
Yep. Yeah.

Miles Ukaoma (19:19.344)
But like where I'm loving this and I type like fairly fast, you know, I think most users are like lower than 70 words per minute. But whisper flow gives you this ability to just speak into your microphone and receive a fully transcribed grammatically correct output from what it is that you were saying.

And you might, you might wonder like, okay, well, why is that of use with, with AI? can just type into the input box. Uh, but what I found, and at least for myself, I don't know if this is, this is true for everyone. What I found is that I provide far more context when I'm speaking into an input.

than when I'm typing because like there's this like mental labor of spelling out the words, thinking about what the syntax should be, the grammatical nature of what it is that you're trying to say that kind of like makes you self edit how much you're putting into the input. But I think that that's like to the detriment of the output that you're receiving because like, are these LLMs feed off of? Context. So when I'm able to just get into the input box and kind of like give a brain dump of what I'm thinking, I think I'm providing

Chris DuBois (20:03.576)
Definitely.

Miles Ukaoma (20:32.147)
providing far more context than I would have if I were just typing into the input box. like, Whisper is like not just been a tool of getting more productivity from myself in terms of like, can talk faster than I can type, but it is also improving the output that I'm getting from AI because I believe that I'm providing it more context of the challenges, the problems I'm trying to solve, what my expectations are, and therefore it's delivering.

closer to that ask than if I were typing and self-editing down based off of just the friction of typing.

Chris DuBois (21:06.862)
Right. All of our training around writing is how do you get a concise message across, right? So that someone actually consumes it quickly. And so when you can just talk, it definitely unlocks something. I'm in the, I can probably do in 60, 60 Guam. I haven't tested for awhile since probably since middle school, but, like probably 60 Guam at least for typing, but I'm looking at my whisper flow and I'm in the top 5 % at like 140 Guam or

Miles Ukaoma (21:13.413)
Right. Yep.

Miles Ukaoma (21:36.475)
Yeah. Yeah.

Chris DuBois (21:36.812)
words per minute. but the, but it just makes it so much easier to be able to go in. What I like about whisper is that it also will, it organizes your thoughts for you. Like it, doesn't just like, I can word vomit. can have my, whatever, it'll clean all those up. And so the output is still something that looks good. but it also, you know, has all the punctuation is correct. And it, like, if I tell it, if I'm giving it a list,

Miles Ukaoma (21:53.457)
Exactly, yeah.

Chris DuBois (22:05.74)
It'll actually put a colon in there and then start the list for me. And so it's like little things like that. but you also have AI built into it, right? So you can highlight whatever texts you just said, and then give it commands to actually clean it up or like, can you actually structure this in this way? And it'll fix it for you. And so like, it makes it even more powerful. Cause now I don't have to go back and like edit every little line. I just tell it what I want. It'll organize that and then I can use it for whatever I need. yeah, I love.

Miles Ukaoma (22:08.571)
Mm-hmm.

Miles Ukaoma (22:31.737)
Exactly. Totally. I love WIPR.

Chris DuBois (22:35.02)
Whisper has been also command mode with it, which I haven't really played with a lot. I need two more, but you can just give it other commands. Like you can access perplexity, right? Claude, all of those other things right from it where just a different button combo on your, on your keyboard. And yeah, super useful. Pro members just got access to their iOS app as well. So I've been using that a bit for like texting and things like that. And it's so helpful.

Miles Ukaoma (22:46.282)
What

Miles Ukaoma (22:52.547)
Okay, now you're teaching me something, that's something new.

Chris DuBois (23:05.324)
So like for, especially when driving or something, like I don't like texting, but if I can just talk to it and stuff, and it, actually know it's going to write down the correct words. It's a.

Miles Ukaoma (23:15.109)
I think that's that. So that's kind of the point here. Like if you can really quickly, cause you've used the iOS version, I haven't used that version yet. Tell us why whisper is a better solution than the native dictation that you have inside of iOS. Cause like that's something that I get asked quite a bit when like I'm telling people about whisper and I'm like, this is the bee's knees. Everybody's just like, we already have dictation. Why do I need to add this additional tool on top of the native dictation that's there in Apple?

Chris DuBois (23:28.792)
Yeah, way better.

Chris DuBois (23:39.981)
Mm-hmm.

Chris DuBois (23:43.746)
Yeah. So I guess iOS is dictation. I've never actually verified this, but I someone else talking about it. I'm assuming it's true. the dictation is like four years out of date. Like they've, they just haven't updated that specifically and stuff. So it's just not going to do as well as a tool that's now optimized with AI to do everything. I will, when driving somewhere, I will often dictate whatever I want into a doc so that I like, I could write a full article while driving somewhere.

Miles Ukaoma (24:13.455)
Right.

Chris DuBois (24:13.484)
Right. Hands, hands are on the wheel. I promise. But the, used to just use like Otter, right? I'd record my audio and then late I'd get back to my computer and then I would start cleaning things up, moving things around. And then I went into just using the dictation tool, like the built in iOS dictation to put something right into a Google doc. But then I have this massive paragraph that I have to go like find the breaks and clean it up. And it's collecting all of my us and ums. Now, when I do it with whisper, it's like, I have an actual article that like,

It takes almost no time. format it a little bit and then I'm ready to go. So yeah, I, I love this. The only awkward part, sorry, I'm going to throw this one in is any of these like tools you talk to, it still feels really weird to do it in public. Like,

Miles Ukaoma (24:46.702)
Amazing.

Set. So the one, okay.

Miles Ukaoma (24:59.659)
yeah, 100%. No, no, no, like it's even worse when you have like young kids that are just like learning language. We have a Google Home. I got these, the Meta smart glasses the other day. how these things work and they all have their own individual commands, you know? And my three-year-old daughter.

Chris DuBois (25:14.683)
yeah, nice. Looking sharp.

Miles Ukaoma (25:24.317)
is like picking up on all of them as well. like she'll be like, Hey, meta, like take a photo. Hey, Google, like play Gracie's quarter. I'm like, this is like they're like real AI natives, but I still feel weird. Like walking around, like I would never be in public saying like, Hey, meta, take a photo of this. Like it's very odd.

Chris DuBois (25:31.278)
Yeah.

Right, they're pulling everything up.

Chris DuBois (25:42.828)
Right. It's weird. My, biggest frustration right now is that one of my dogs is his name is a Deerigo, which is the main state motto. But for short, we call him Deery. But every time he's outside and like waiting to come in, I'll be like, Hey Deery. And then my phone's like, Hmm. Like what? I'm like, no, I'm not talking to you. And then my kids will yell that too. He's not talking to you, Siri.

Miles Ukaoma (25:55.675)
He.

Siri. Yeah. Go ahead. I'm not talking to you, Siri.

You

Chris DuBois (26:08.994)
Yeah, okay. What's the next tool? What do we got? All right.

Miles Ukaoma (26:10.124)
Um, yeah, I got one more tool and I think that this tool is, it's a really good segue for our discussion as well, because we're branching out from these just knowledge-based tools. Like that's what we've been, a lot of people use like chat GPT and, and LLMs in general as just like a knowledge-based tool. They want to query it and get a result. And what we're seeing or hearing a lot of inside of 2025 is the role of agents.

and how agents are going to come into the foray, how all the opportunities exist behind this advent of agents. So an agent is just a LLM that has access to a tool to take action on behalf of the user. So that's how I'm defining agent, and that's how I see most people broadly using agent as well. So what's the so what of all of this?

Whereas today, like you just go into a chat GPT and you can ask it a question on like who won the gold medal in the hundred meters last year at the Tokyo Olympics or something along those lines. It's going to be able to search Google and pull down that query if it's not already and it's knowledge and give you the answer to it. But like there's not really the ability to take action after that. So if I wanted to be like.

write this individual a fan letter and send it to the email that you are able to find for them. The LLM is gonna be like, I don't have access to any of those tools to take that action on your behalf. And that's where agents come in. Whenever we have a piece of knowledge that we have and we want to take action on it, we have to be able to hand it over to these agents. And there's two methods in which people are going about building agents today.

one in which I am, I'm, have like a, not a vested interest, but I truly think that it's going to be the. leader in the landscape as things continue to progress and the other, which is already established and has uses, today. So the one that is established and has uses today is what's called function calling. And this is where, if you're working inside of like an open AI, their platform, their dev tool, if you're working inside of anthropics, Claude, their, their console log.

Miles Ukaoma (28:32.597)
You have the ability to define certain functions or tools that are just existing services that you're already taking advantage of. Right. So Slack can be a function or a tool that this AI agent calls. And inside of the Slack function, we can define all kinds of activities. I can draft new messages. I can search messages from existing channels. I can create channels. I can edit.

These are all functions that you have within Slack. When you're interfacing within the platform, you can click around in the GUI and do all of those things. However, there's also an API that can be exposed to these LLMs that have all of those services defined within them as well. So what I'm seeing more of on the tool side is tools like Raycast, which is, you know, become my favorite tool.

of the year so far. Raycast is a spotlight. If you're on Mac or like an Alfred, it's a website launch alternative. And it gives you a lot of shortcuts that are built into your Mac OS device, right? I can, for instance, at Slack and say, send Chris DuBois a message for at one o'clock. So here in like another 20 minutes.

around X, Y, or Z topic, and it will schedule that message. And at the time that the threshold has been hit, will fire it off using one of these services based upon the tools or the functions that we've defined for the LLM. It's like, that's one path that leads to the same result. It's just an agent, an LLM that's able to take action on our behalf. And then the other path, and then we can, I'll shut up so we can discuss these things, is MCP.

or a model context protocol. this is like, this is a game changer. I know everybody says that, like they're like, it's a game changer. But I truly believe that it's a game changer because it creates a standard around how these LLMs communicate with all of the different services that are out there. So that first example that I shared with you, it's pretty intensive from a development standpoint. You need to be familiar with APIs.

Miles Ukaoma (30:55.867)
You need to be able to sit down and kind of think through what are the different functions and where should the LLM be handing off to the individual tools. And it's very dev intensive task. And it's also not portable. And what I mean by that is if you define all of these tools for open AI and the models that exist inside of open AI, you can't easily take them and start using them with Claude or with Gemini or with any of the other frontier.

models, you're to have to rework everything in that instance of like the first tools and function calling. Whereas MCP, because it gives us that defined standard of how these systems communicate with one another, it allows us to reuse these components that we've built across all of the different LLMs. So you can have a MCP server that is a Slack MCP server and has all of those functions.

that are defined within the send, how to edit, delete, how to search the channel. And I can take my LLM and show it, hey, this is an MCP server that is used for any time you need to interface with Slack or the user asks you to interface with Slack. And I can have it take actions on my behalf through the MCP server. And it's agnostic of a certain platform. Claude.

ChatGPT, Gemini, all of them can interact with this protocol. So I think that that's a major opportunity that exists for agencies and selfishly for myself. I'm using them today in Raycast. And I think that like...

That is the natural progression of where we go with LLM. It's no longer just querying knowledge. It's also, let me query this knowledge so that I can take an action that drives some type of value for me or my client or my agency.

Chris DuBois (32:57.326)
Yeah. With, with the MCP, um, server is that so like certain applications can make it for themselves, right? As like an API kind of call similar to like, as if you were just calling that that way. Um, but for would agencies need to kind of make their own structure for like, for their LLMs to be able to talk to it directly, or is this something that just exists out there? wait for.

You know, slack to make their own MCP. And then you reference that.

Miles Ukaoma (33:27.621)
Right. So that's, that's why I think that this is like amazing. It like entropic created the protocol, but they're not the main creator of all of the different MCP servers. Like what's happening is exactly what you just said, Chris providers are creating their own MCP servers. So like, what's likely to happen is like in another month, you're to see Slack come out with their own MCP server and now everybody can just go and like, just like an API, they can go and they can just consume from the services. that

Chris DuBois (33:52.684)
Ray, open API.

Gotcha.

Miles Ukaoma (33:56.741)
the LLM is able to plug into it. It's just like a defined protocol. So providers are going to start creating MCP servers. And additionally, there are MCP server catalogs that exist today. A really popular one is smitheree.ai. Smitheree.ai is just a compilation of all of these MCP servers that have been created and are maintained by providers, other agencies.

As well, like to your point, like there's going to be proprietary data that your clients have, right? Maybe it's living inside of Tableau. Maybe it's living inside of some other data lake and you could create an MCP server, which has access to all of that data. And it communicates securely across the, the, the client's proprietary data, which we know is like super important to get a unique output or a relevant output for what you're working on.

It's like if you're an agency thinking through, okay, what are the MCP servers that my client might have needs in because they're starting to adopt these LLMs more and more. And they're probably going to want them to take action on their behalf. That's going to save somebody time. It's going to make somebody more productive inside of the business. So like, what are those MCP servers that we can go and build? Those are the opportunities that exist for.

agencies. But as practitioners, it's like, how do I start leveraging these MCP servers that are getting put out by different providers like Slack and the like?

Chris DuBois (35:25.122)
So I'm seeing an immediate fix. Say we use Tableau tracking all client results. Every test that I run for content, for CRO, everything goes into Tableau. And I can now use this. This basically telling LLM, hey, my client where we're seeing a bounce rate of 60 % on the homepage, what are some tests that I can run? And it'll cross-reference every test you've ever ran that was successful and give you immediate tips on how to do this. Now you can just go execute.

Miles Ukaoma (35:45.605)
Exactly.

Chris DuBois (35:53.486)
So this is, this is like the perfect example of what I'm, what I was bringing up earlier. Where it's like, we don't have to now think through all of those steps. Obviously you still want to understand why and how all of this works and stuff like, but you could potentially bring on a new hire and get them trained up really quick. So they start seeing those patterns and learning the patterns as they're executing, but they're using the entire knowledge base that your agency has developed. Like just so much more power they can go into, to an agency.

Miles Ukaoma (35:53.604)
Exactly.

Miles Ukaoma (36:20.686)
Yeah. And, and like, I think we all know what was the limitation before, because it's not as if like people weren't doing these things. We were, we were taking the time to create that proprietary data and we would, we would get like Google sheets of all of our personas and then we would turn them into JSON and we would feed the JSON into the LLMs, but like maintaining the data became a whole job.

Like we had to have people dedicated to, well, this client has updated this persona. Okay. Now we need to go through the process of updating the Google sheet and turning it into the JSON and uploading it into the vector store and notifying the client. And like the next thing you know, like 10 steps later, it's like, okay, it's updated for you. And now it's like, no, it's just the MCP server. There's whatever's happening in Tableau. It's able to consume from, and we have one MCP server that we've defined for these certain tools. And it just speaks to that.

Chris DuBois (37:12.974)
Mm-hmm.

Miles Ukaoma (37:13.201)
Like there's not that whole build step that we were having to do in the past years.

Chris DuBois (37:18.414)
Right. There's even more, I think, flexibility for human error with it, where it's like, if one person writes something differently, right. Then it's okay. Like this is a simple mistake. And I say this jokingly for anyone from Mississippi, but whenever, whenever I was like onboarding people to HubSpot and we were talking about how to set up different fields, the difference between a single line text field and like a dropdown, I would, if we were talking States, I would give them the reason is like,

You want someone to actually be able to choose the exact option because they might have 10 different ways to write something right. Like Mississippi, where it's like, can use the acronym, you use miss Mississippi. You can spell it with one S two S's five S's like, cause people from Mississippi probably spell it wrong. And that's the joke, but one client was actually from Mississippi and wasn't thrilled. Yeah. So it's a joke people get over it. but anyways, and so like reporting was a nightmare, right? If you have a single line text.

Miles Ukaoma (37:50.895)
Yeah.

Miles Ukaoma (38:02.865)
If you're from Mississippi, I'm sorry in advance.

Chris DuBois (38:14.926)
And so, but like with something like this, even if I'm building out all of our in Tableau, I spell something differently, like you can, the AI can actually read that and use some logic behind it and say, Oh, it probably means this. And so now it's okay that the new hire didn't know that, you know, we always put input this exactly like this. Um, so, well, yeah, so that's, that's a pretty cool one with, uh, going back to Raycast. Uh, so you only told me about this yesterday.

And I immediately downloaded it and started using it. But it's pretty sweet so far. And so I'm excited to actually get into it.

Miles Ukaoma (38:49.509)
What are you noticing first? Like what are some of the...

Chris DuBois (38:52.366)
I'm just using it to like pull up, pull up different notes that I've had and stuff. So I don't have to go search for them. can just tell it to like, Hey, can you find me the note on this? And it's like now, instead of spending, even if it's only two minutes, it's like, can just tell it what I need and you can use the voice dictation with it as well. And so I can just say, Hey, find this for me. And then I start finding that while I'm still doing whatever else I need to be doing. And so my, my actual like speed to, getting stuff done is way better.

Miles Ukaoma (39:11.729)
doing what you're doing.

Miles Ukaoma (39:18.073)
One other thing that think that like agencies are going to love is Raycast has this video downloader tool that works with YouTube. And like, I never use it to download videos, but I do use it to grab the transcript from the video. like, you know, there's lots of content that I want to consume, but I don't want to sit through three hours in order to get like the 10 minutes of value that's embedded within it. And now I can just feed the transcript.

Chris DuBois (39:25.219)
Mm-hmm.

Miles Ukaoma (39:45.733)
And I know the topic that I was hoping to get from the video and just query that transcript with like, pull out this very specific topic that I'm looking for. What were all the insights that were said about it? And it's just going to distill that three hour long video into the 10 minute snippet of value that is relevant to me.

Chris DuBois (40:04.258)
Right. Which for content creators listening, that's more reason to, to provide as much value as you can within your content Like people are going to start doing things like this. They're not going to sit there and watch all of your content. They're just going to pull the highlights. And if AI just pulls out whatever it thinks the highlights are, and they're not very interesting, people are going to stop looking at your content period. and so like the way that people are going to start consuming some of these things is going to change, or just get really entertaining.

so that people want to watch for the entertainment side and collect those insights as they come. But I got two AI things that I wanna add to this list. So the first one being ChatGPT's voice feature. I am loving this. So you can just have a conversation with ChatGPT. I will say the one thing you should do at the start.

Miles Ukaoma (40:45.873)
Please.

Miles Ukaoma (40:50.897)
I'm not using it. Tell me more about it.

Chris DuBois (41:01.346)
give it like a written instruction that says, do not summarize what I am like my answer to everything. just give me like whatever the next question or whatever, like whatever the conversation is about, like just continue the conversation. Otherwise it'll say, Hey, interesting thoughts about this. Like, I like how you said this, this it's like, no, I know I just said that. Right. Let's move on. And so with this, I'm, I'm a talker outer sometimes, like as a solopreneur, I don't, can't just pull someone onto a zoom and be like, Hey,

What about this idea? And so when I want to flush something out, I'll just walk around my office, like throw in my AirPods, start talking to chat GPT. And I can just say, keep, I want you to continuously just help me get to the heart of what I'm trying to figure out here, whether it's for content, a new offer, like anything. And I'll just start talking about it and it'll keep poking holes in whatever I'm thinking. And I'll just keep going deeper and deeper. then at the end I say, okay,

provide a a summarize, a written summarized list of everything we discussed. And then it'll say, yep, here you go. Good to go. And then when I go back into chat, GPT, I can just copy and paste this entire summary of everything we just got to. And just the fact that like me working through this is like, I got so much more done, but I hate going for long drives. Like I just, because I can't do anything. I am not productive.

Miles Ukaoma (42:14.491)
Right.

Chris DuBois (42:22.69)
I would rather fly. I hate flying, but I'd rather fly somewhere because I can at least pull up my laptop and do something. this past weekend, I had to drive down to Rhode Island for drill weekend with the army. And I spent over an hour talking to chat GBT in different threads to like work out article ideas. I was having different like business ideas, trying working through some problems my clients are having so that I can get them a better answer. And, I spent an hour like just consolidating all of these ideas. So when I got to my destination,

I had all these different threads that I was able to just pull in. sent some Slack messages, right? Like for people that needed them, I moved the different article ideas where they had to go. And it's like, I was actually like super productive on this car trip. And so, yeah.

Miles Ukaoma (43:06.193)
So I feel, I feel like it's a continuation to what we were discussing. Like my prediction by the end of this year, that voice mode also is an agent. So rather than you having to like, take the mental note of when I get to my destination, I'm going to send off this Slack message, or I'm going to send this email or we're going to notify X, Y, or Z client. Like you'll just be on your drive and you'll say, okay, like now that we've arrived at a solution for this problem, update the client and it will fire off the email on your behalf.

Chris DuBois (43:17.219)
Yeah.

Chris DuBois (43:34.722)
Yep. And that's going to be so much better. last tool I want to bring up lovable.a.dev whatever. So this whole vibe coding thing, I didn't know that's what it was called. People started talking about vibe coding. I'm like, what is that? but I realized I've been doing this. And so I'm a, like an OG vibe coder, I would say, no, but I've just been playing with it for the dynamic agency community with like, we have people who could.

Miles Ukaoma (43:41.967)
I love it. Yeah. Live coding. Right. Right.

Miles Ukaoma (43:56.059)
Yeah.

Chris DuBois (44:02.978)
could get some value from different resources, different things like that. And I'm like, I could just build some of these things. And so like for one, I made a database for any members of the community can go into this, add themselves as a, like as a member. And so it's a directory essentially, right? But they can add themselves into this and update like, if you have the question around the topic, here's what you should ask me. So now, cause we're, you know, as the community is growing, my

My whole goal for the community is anyone can come in here, any agency owner or leader can come in and say, I need to know, or like, I have a question about this. Who do I talk to? Like you can go to the channel and just say, Hey, anyone, you know this, or you can actually ping the people who in this database have pointed out like positioning, talk to me, right? For sales, talk to this person. And anyone can go in and put that up there, but they can also give them another way to highlight their agency, like, or their business, like a little bio and.

I built all of it so that members can add themselves. have my own little admin dashboard. can go and play with everything. And it was easy. Like it really, like I had to go back and forth with it a couple of times to get exactly what I wanted, but me with no coding experience whatsoever. Right. You've seen some of the stuff I've tried to do. Like it's, right. It's like the Chris test with sprocket rocket, like, he can figure it out. It's probably fine. but like, I'm.

Miles Ukaoma (45:21.093)
Ha!

Chris DuBois (45:31.64)
coding these little tools that are super even one I connected. so I made a, a niche assistant tool. I would need to give it a better name, but essentially an agency can come in. It was a custom GPT where they can come in and just, it'll prompt them the questions to help them figure out what niche are they actually serving what industry sub industry horizontal vertical of like, who's the job function that we're actually working towards and then what problems that.

And then it would do the math and say like, okay, well, based on what the, bureau of labor and statistics says, like this industry is growing. Here's what, you know, the competition in this space and it'll do all that for them. So they can know, okay, I'm on the right track. I took that and just put it into like an actual platform. So now you don't have to go to chat GPT to do this. You can go into like a sync, a website, put the inputs and it'll give you the whole output right after. And so it's like, it's almost a cleaner interface for someone to be able to, get the results.

It's like, I'm loving this thing.

Miles Ukaoma (46:29.777)
100%. Now, I mean, these tools, honestly, we could do a full episode. Maybe we do that at some point. Let's do like a full episode because I have like way too much to say on this. Like, I agree. You're probably like one of the earlier vibe coders. I feel like I was the legacy vibe coder because like I was vibe coding back when you had to go to Stack Overflow and copy the snippets.

Chris DuBois (46:55.086)
I don't even know what that is, so.

Miles Ukaoma (46:56.185)
Like exactly like that used to be what developers did before chat GPT existed. Like it's not as if we were all writing this stuff from scratch. We would go to stack overflow and find that really smart dev that had done it the correct way before. And we would just.

Chris DuBois (47:09.037)
Gotcha.

Miles Ukaoma (47:09.937)
you know, steal the steal. would, we would borrow the majority of what they have written and adapt it for our, our use cases. You know, there's this meme that's going around where it shows somebody that's just made a latte and has a nice like heart design at the top of the latte. And there's a coffee, just a black coffee sitting next to it. And they're like, this is pretty much coding or vibe coding. Somebody took a spoon, grabbed the, the, the, the heart and dropped it into the other thing. And it's like, see like.

Chris DuBois (47:34.796)
Yeah, I have seen.

Miles Ukaoma (47:39.891)
Look at my code, guys. But this is the direction that all of this stuff is going. Last thing I'll say on this is there's 150 million developers active on GitHub today. So we can use it as a proxy for how many developers there are in the world. Maybe there's 200 million of them. These tools are going to democratize the creation process. You're not going to need to have learned formally HTML, CSS, JavaScript, Python.

C class like all of those languages are still going to be important and there's going to be lots of AI augmented developers, but there's going to be a whole other rung that opens that comes in based off of this vibe coding. Like we're to see that 150 million go to 250 million and 350 million and half a half a billion like in short order.

Because people are going to do more of what you're doing. Like lovable is not the only one. You have lovable, have Bolt, have Versel, you have Replet, have their cursor, have tailwinds. You have so many of these tools that have been pushed out in the last year with the express purpose of democratizing.

Chris DuBois (48:38.124)
Yeah, cursor. Yeah.

Miles Ukaoma (48:49.233)
the development process, which I think is going to be a net positive for all of us. It's going to open up a ton of opportunity and we're going to see way more development taking place. Makes me excited as a dev, but yeah, Vive coding is, it's a lot like the coffee. Like we're just like, look at my code.

Chris DuBois (49:04.898)
Yep. All right. I'm, I'm serving someone a black coffee that looks like a latte. So, all right, we got to wrap up here. two questions before you go. What book do you recommend every agency owner should read?

Miles Ukaoma (49:18.321)
Cal Newport deep work. We do not, you know, we've got to be able to work deep and like a lot of things that we're doing, especially with like leveraging these AI tools come off of the backs of that deep work. So I think Cal Newport's deep work is like my favorite book and I recommend all agency owners and agencies give that a peruse.

Chris DuBois (49:22.51)
Classic one.

Chris DuBois (49:40.15)
Every time someone brings up that book, I was listening to it during one of those like crazy workouts we were doing. Like it would just tons of jump lunges and everything. Like one of those challenges at the, like the agency. And, now every time someone mentions that, I just think of jump lunges and being in my basement suffering. so, anyways, no one needed to know that. last question, where can people find you?

Miles Ukaoma (50:03.749)
you guys could find me at milesycoma.com.

Chris DuBois (50:44.206)
Yeah, sure.

Chris DuBois (50:50.37)
Yeah, let's actually, I think we'll do more of these in the future if you're down for it. Awesome. All right, well, thanks for joining.

Miles Ukaoma (50:57.563)
Cool, I'm down.

Alright man, enjoy the rest of your weekend. speak soon.