Leap to Scale
Leap to Scale is for technology curious leaders of service firms who want higher margins without adding headcount. Each week, Justin Davis and Greg Ross-Munro show how to turn firm expertise into repeatable, sellable technology products: SaaS, packaged workflows, and AI-powered tools clients can buy again and again. With AI, more of your know-how can be captured, standardized, and protected as IP instead of being rebuilt in every engagement.
We focus on practical decisions: what to productize, how to price it, when to build vs. buy, and how to use AI responsibly without risking delivery quality or margin. Clear steps, real tradeoffs, usable examples.
Leap to Scale
Hiring Your First AI Agent
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
AI is changing how companies hire, but not in the way most people think. In this episode, Justin Davis and Greg Ross-Munro unpack why AI agents are replacing tasks, not entire jobs, and what that means for service firms trying to scale efficiently. They explore how modern AI workflows can automate reporting, proposals, market research, and other repeatable operational work using tools like Make.com, N8N, and GPTs. The conversation also covers where non-technical teams can go surprisingly far on their own, and when it’s time to bring in software professionals to productionize and secure these systems for real client use.
Okay, wait, wait. So they they Claude changed the name three times.
SPEAKER_04What oh it was Claude Bot. So what it was is there was this there's this new agentic framework um called Claude that was called Claude Bot. C-L-A-W-D bot like lobster Claude.
SPEAKER_03Unrelated unrelated to the Anthropic?
SPEAKER_04Yeah, it was meant to be. They admitted it was meant to be a pun. It was meant to be a player. It was supposed to be cute. Right. Well, um this thing gets sued then. Yes. Well, this thing rocketed in popularity um in an unexpected way. Um ended up with like 70,000 uh favorites on GitHub, which probably equates to like 10 XSAT actual users, you know. Um, and so um they obviously got the cease and desist, and so they renamed it to Moltbot, M O L T Moltbot, which that's just a terrible name. That's just not a I get what they're going for, lobster spool. And then um, and then now they have renamed it to open claw. So um that is three name changes when you have really like hundreds of thousands of people looking at you. So hey, you can do anything in tech.
SPEAKER_01Crazy, weird, weird thing to do in a business.
SPEAKER_04Hey, welcome back in here to Leap to Scale. This is the podcast that helps service companies leap to scale without adding head count. That is, add add revenue without adding people to the picture. Um, you do that with technology and other things. We're gonna talk about it here on this show. And today, with my co-host here.
SPEAKER_01Hi guys, uh, I'm Greg Rossmanroe, and my co-host is uh I'm Justin Ibens.
SPEAKER_04I did not say my name. I don't think I did. I don't know. It's it's I think we're both a little brain dead today. We are a little bit brain dead today. But it's okay. Um But we're gonna we're gonna have a fun conversation today because we're gonna we're gonna talk about hiring in 2026, and I think that hiring in 2026 is different than hiring in 2025.
SPEAKER_00Justin, you just said that this podcast is about growing your business without adding it.
SPEAKER_04Now you're talking about hiring. Yeah, that is a little bit strange. Maybe I should unpack that. We're not talking about uh hiring heads, we're talking about hiring bots. I wish I had like a really cute, like I I really wanted that to like end on a zinger.
SPEAKER_01I think we were trying we were trying really hard to set that up and we were trying to help. You it's this is this is how you start a podcast is by being cheesy, and then you can you slowly unpack the introduction into actual substance, which hopefully is what's coming next. Yeah. Well we will all find out together. Um you say bots. I said bots, but I think the um the we're probably gonna name this this episode something like how to hire hiring your first agent, your first AI agent. Yeah, something like that. Can you for what is an AI agent, Justin?
SPEAKER_04What's an agent? What's a bot? What is all of this, right? What is the difference between that stuff? When we talk about agent, what we're talking about is something that's powered by AI, just like ChatGPT that you've used before to answer simple questions that you might otherwise Google, or you have it write you a song, or you have it, you know, help you with a work problem or something like that. Um it is imagine putting that kind of on a loop and just letting it run forever. And it just always responds to you and can just do work for you and can just continue doing tasks kind of autonomously over longer periods of time. What we would call like multi-turn tasks, right? So, not like kind of a I hand you a thing and then you do it and you give it back to me and we're done, not a single exchange, but a kind of multi-turn exchange where it might be iterating on its own, it might be colling resources in that it needs as well. And so agents typically have tools. So it's sort of like a robot version, well, a virtual robot version of like a chat GPT is a way to kind of think about it.
SPEAKER_01So if I had my own little ChatGPT or something like imagine it's in a little box and it's listening for something and waiting for something to happen occasionally, and if that thing happens, it does an action, or it's on a timer. So like every every 20 minutes or every whatever, it goes and it does a peer a set of actions. Is that is that like a way of looking at it?
SPEAKER_04Fantastic example. I'll give you a great example. I have something, I actually have an agent uh running for me right now. Um and what one task that this agent has is I have a mobile app that is in the app store. Um, and I want to know every day information about like what happened yesterday in that app, how many people signed up, how many people did certain actions, that kind of stuff. And so I had my agent just send me a report every morning of the key things that I want to know. And so now what it does is at nine o'clock every morning, I did I get a DM from this thing called Moltbot still that tells me, hey, here's how many new users you had, here's what the engagement looked like, here's what people are doing, and that, and I could even extend that. I could ask it to even give me recommendations for the day, give me some strategic things to think about, right? All of that. And so it starts to become, yeah, this kind of autonomous helper that way. And you're right, it can either be something that is sort of there that you can just talk to anytime that you want, it can also extend to being something that is something that you might trigger to do kind of what I would refer to as like medium or long chain automations, which is a multi-step kind of automation instead of say like converting a file from one file to another, it's a short chain automation, it's like a one, one or two-step thing. But these kind of when you get into multi-like medium and long chain uh automations, generally speaking, we're talking about agents doing it at that point.
SPEAKER_01Yeah, I I um I have an agent that uh I have an uh agent that runs every single night at the at the moment, just to give you an idea, uh, which is a uh competitive um market analysis agent. And it looks at um a few of the places that that I'm interested in playing in the world of business. And there's only there's a small number of competitors, like let's say three or four real competitors in that space, and I don't have time to look at their websites every day. And so I just have an agent that monitors their three websites and their blog posts. And you know, you could like do a Google post like this, but every time, but maybe nothing that much changes. So every single day, these agents go look at those websites, they see if there's any news about those companies, anything that they've added or updated on their website, and then it uses the intelligence of the large language model to be like, is this like something that Greg should know about? And then if it does, it sends me a Slack message about it. Yeah, that's a great example. It's a great example. It's not really like if if we're we're talking about like hiring an agent, right? I would never have gone and hired a full-time market research person. Maybe I would though hire like a part-time like or a VA.
SPEAKER_04And I think I think when you start talking about like tasks that VAs might do for you, like a virtual assistant. Virtual assistant. Um are the types of tasks that you right, because you might take say to a virtual assistant, hey, every morning I want to have this quick report on my desk, right? Uh, and they'll take a couple hours to do it or whatever every day, and they're working across multiple people or multiple tasks or whatever it is, not full-time just on that one thing. Yeah. Sure.
SPEAKER_01Yeah. Yeah. Where did you hear the thing about that jobs are not the smallest atomic group of work? Yeah.
SPEAKER_04Right. Yeah. So I was listening to a podcast or I saw a clip of a podcast. Mark Andreessen was talking to some on Twitter.
SPEAKER_01Mark Andreessen, okay.
SPEAKER_04Yeah. And he made the comment that um when he was talking about like people talk about like AI replacing jobs, AI replacing jobs. And that that's not really the right way to think about it because uh the job is not the atomic unit of work, but the atomic unit of work is actually a task, right? And that jobs are just bundles of tasks, which if as soon as you hear that, that's probably going to be that's gonna make a lot of sense, right? Um, and so what AI is doing is AI is automating tasks. And so it's not automating all of the tasks in the bundle. It might just be automating some of the tasks in the bundle, the tasks that are left that a human still has to do, like going and sitting down and taking a prospect to lunch or go, you know, doing your account management type of role, that these other things that really need human uh taste and judgment and interaction put into them, those things stay in the human bundle. But a lot of these other tasks can get kind of extricated out. And now what you've done is you've taken one job, you've actually created a two. You've redefined the original job by its new bundle of tasks, and it might get other new ones put into it. And also this bundle that you've extricated for the agent or for this other thing is a new job, and that could be taken by an agent.
SPEAKER_01Yeah. So the the thing that uh that I've done that is like reasonably powerful is um is stringing these agents together, right? Is having these little individual like a one agent does one task. Like it goes and it looks at the website and it gathers data every single day. But then I have another one that I build later, right, that will do some sort of analysis on that thing. Right. And then maybe I have, and then the output of that analysis, I have another little like thing that I build that will take that and put it all together in a report and summarize it all for me, kind of thing. Like the maybe not the greatest example of of because these are just purely informational. And so like these things seem to a lot more things, just informational work. They can actually reach out, and we can talk about that here in a minute. But um uh string those together in a workflow, now I've taken to go back to this idea of like tasks being the atomic, the smallest atomic grouping of a job. And if I string these tasks together, have I actually built have I have I built something that could that is a job? Would you would you argue that?
SPEAKER_04Yeah, I think I think you could get to that argument. Absolutely. I I think that if you think about uh an agent that has a set of tools available to it that and each one of those tools corresponds to a task, then I think that you could you could make an argument that you have created a job like structure for which this uh this agent is doing.
SPEAKER_03Yeah. Yeah. I think that's a good thing.
SPEAKER_04So that's uh I've never really thought about it. It's interesting.
SPEAKER_01Yeah. Yeah. Okay, so if if if if I've got I'm in my Slack all day on my Microsoft Teams, whatever, and there's a person inside of my organization whose job it is every morning or every Monday morning to drop some sort of report in a Microsoft Teams channel, uh, is that person's job on the line?
SPEAKER_04Right, no. Right, but that task that they are doing might not be needed anymore. They may not need to upload this report because we have ways to do this with technology that it's just easier. But that individual now, if if all that person's job is, is a single task, and that is the only task that they do, then yes, I I yes, correct. Okay, but it's unlikely that that is the case.
SPEAKER_01Yeah, it's very unlikely. And there's probably more important things that they should be doing that are human-oriented things anyway than pulling spreadsheets out. Okay, so then if I want to start, let's say if I if I didn't know anything about this and I want to build my first, I want to hire my first agent, because this is kind of like the jobs to be done kind of problem in an automated way.
SPEAKER_00I don't know. Yeah, that's interesting.
SPEAKER_03Yeah, it's interesting to a point.
SPEAKER_01I don't know. I didn't think about that and roll it around my head. But like, okay, where do I I know how to use ChatGPT? Let's say.
SPEAKER_03Yeah.
SPEAKER_01What and that's it. I I I talk to ChatGPT, I'm like, hey, here's a PDF file, give me a report. Great. How do I go from that, level understanding, to I'm building an agent that runs some somewhere in the background all the time. What's your like what is your uh we have to hire not just the agent, we have to hire like the whole stack, right?
SPEAKER_04Yeah, that's right. That's right. Um, and so yeah, you you you do kind of have to hire the whole stack. It's almost like hiring a contractor, right? That comes with a with talent kind of in it, you know, um, in a way. I I think like so if the question is like sort of where do I go? How do I how do I kind of cross that bridge? This sounds interesting. What what would you do? There's a few parts to it. I think the first thing is to get clear on what it is you want this thing to do. I think the first thing to do is to actually, I would recommend sort of taking an exercise and maybe write at the job description. The job description is the bundle of tasks. Maybe it just has the one task, and then maybe it's only just a single job, it's a reporter, it's fine because you can have 10,000 of these, right? And it doesn't cost you much marginal cost. So you can have now specialists who only do one thing, but do that one thing very deeply instead of and so that some other advantages that we won't talk about right now. But um, either way, write up the job description. What do you expect it to do? Who is doing it in the organization today? What are their inputs and what are their outputs? Look at that, right? Because these are going to define how you need to create this thing. There's it's not like you sort of go to like aiagents.com and sort of get an agent and it comes in a box and you set it up on the on the bookshelf and it sort of starts to do work for you. The idea of an AI agent is is a concept. It's a it's a pattern, if you will. And so an agent might be, might take the form of going to something like make.com, which allows you to very easily visually just build little simple workflows that says monitor this Dropbox folder. And whenever they drop the CSV of data in there, take it and send it to this LLM and have this LLM and LLM is like ChatGPT or something like that, have it make some charts out of it and send it to the Slack channel for me. Right. So that's a job. That's a task, right?
SPEAKER_01And so you might build-I think I think we I think hang on there a second, because like there's an interesting almost like definition there of an agent, which is because some people might not be ready for like, oh, I don't even know what's possible, right? I yeah, like, yeah, there's all these like little jobs that I all these tasks that I don't want to do in my day-to-day life. But I mean, I think that just the two things you said there. My suggestion for somebody would be go play with go get an N8N or that's N8N um or a make.com account. Um which is basically just like a way of like building a workflow. Chat GPT, if you use chat GPT, that's the most common one on the out there. Do a work out how to build a GPT yourself, which is by the way, still the dumbest name for what they're they're called. Those assists they used to be called assistants or whatever. Yeah. Find out how to make a GPT and get make.com to trigger a GPT. It's pretty pretty easy. Like you probably teach yourself in an hour to how to do it. And once you've done that, you will have built your first agent. I think that's like a that's a really interesting, yeah.
SPEAKER_04It's a really interesting place to start. I bet.
SPEAKER_01And it doesn't have to do anything, just like just go and experiment with it because as soon as you play with the tools, I think you'll be able to see, I think some people will be able to see what they they could do with this stuff.
SPEAKER_04Yeah, when you go into one of those tools like make.com and you start to look at that, you'll you'll kind of open the palette of things, kind of explore, and you'll go, oh wait, I can input from I oh, it can connect to my AutoCAD. It could connect to my uh my bank. Um, oh, well then if it can do that, then could it help me know this? And the answer is probably yes. And so then it's a matter of like how do you start to break that that thing down into a task that you can define in one of these builders.
SPEAKER_01Uh what's funny is uh AutoCAD already connects to your bank account and it uh just takes subscription dollars out and gives them to Autodesk. That is what that is what it's designed for. We've had agents at software forever. They've just been taking money from us. Autodesk, you used to print money. Yes.
SPEAKER_04Um, I think like that that's a great place to start. Um, you know, you can look at things like ClaudBot. Honestly, here's the thing about this: this stuff changes real fast, and things that get popular, it's very fast to build products now. And so it is likely by the time you listen to this show, unless we release it next week, it is very possible that there are now tools that are uh out there like ClaudeBot that will allow you to do this without needing to be comfortable using like a command line, like a programmer. Um, but if you are comfortable in those environments or willing to learn, things like that, I think are offering a really early glimpse into what this world is going to look like. And I think it's gonna happen really fast. Yeah.
SPEAKER_01Yeah. I mean, uh getting used to those tools, getting used to just a couple of them to build some awareness of how they work, I think is a really good first step for people. Because if if you can start by building those little kind of simple tools, you can string them together to make really, really powerful stuff.
SPEAKER_03Yes.
SPEAKER_01Um, like the one that the one that we always talk about is uh we record the first you you you I don't know, like you have an intake process. You record the meeting with the client as they uh as part of the intake process. You have an agent that has every single proposal that you've ever written and knows how you write proposals, and you have another agent that has um all the questions that you normally ask clients, and then you have another little agent that says, okay, ex listen to this recording, extract out the text, and then what you do is you you chain them together. So you take the text out of the recording, you make you you uh give it the questions and you say, Hey, here's all the questions, and here's the answers. And then based on that question and answer, hand it to the little agent that has seen all of your proposals and has like a set of text instructions on how to build that. And now you've got an end-to-end workflow, a agentic workflow that takes a recording, a transcript and actually puts out a proposal that is like that's a job. That's a job. That's a job. All those thoughts strung together. And then that's that that's a fairly basic one that I've just described. That's not even like something that goes and uses tools and uh that's just got some little additional color in it from the you know, like from the retrieval, uh, the augmented retrieval that it's doing. It's got like a data data store. But it's not doing anything fancy.
SPEAKER_04Right. No, and that, yeah. And that's a very possible thing to do. It's a great example because it it is one of those things that I think you could listen to that and you could go, there's no way that I could go from a conversation and then press a button and have a proposal done. Yep. How is that sounds like voodoo magic, but the word but I think what is interesting about this process, and and maybe what a lot of people sort of learn and start to think about things uh through this is that um when you start to take a job like that, or and you start to break it down into its component tasks, into the bundle of tasks, and you take the task and break the task down to steps, right? And they and you keep breaking it down, you start to realize that it is actually, you can't actually cobble it back together on a lot of these tools. It does take some time and it does take some work and it takes some experimentation, but we're gonna see more and more of this. And and more and more spots on the org chart, when you start to think about hiring in 2026, 2027, 2028, there's gonna be more spots on the org chart that you look at and you say, is there a human or is there an agent in that role? I think that's gonna be a really interesting shift.
SPEAKER_01That's an interesting way of looking at it to seeing like in the situation where you're like working out what the tasks that are involved in a job. Normally when I think of a job description, I think about maybe I probably kind of think about like outcomes, really, right? Like what do I want the outcomes to be? And I don't often think about it in the smaller breakdown of what are the tasks that I want this person to do. I'm like, oh, I want the outcome, the I want these outcomes. So, and then I I try and give the person some level of the human being, some level of freedom on how they achieve those outcomes, right? But with something like this, I it's you you you kind of have to work out what the outc what the outcomes are the same as you would with any job description. But then you've got to like break that down into the atomic tasks and say, okay, if is if there's a step of processes and procedures that I can string together, then I can actually replace that, or I don't have to hire that, or I can free up an existing employee to go and do something else. Um But uh one thing that I've noticed, we've we've we've been doing this quite a bit where people have started to like. Like not the people who are like somewhat technical, but not software engineers. They've been building these things on their own. Right? But there's a certain level where they they get to, where they become. I don't know where where when's the how far can you go as a as a non-programmer these days with with some of these things?
SPEAKER_04Yeah, I mean, I still I think right now, I think you can you can do an impressive amount. You cannot, I there are still some things that are gonna require some specialist knowledge to really be able to do at scale. I mean, we you know, like take an example of if you wanted to to create kind of like maybe customer ports or something on a client level basis or something like that that was quite complicated and involved connecting to a bunch of different data sources and doing a bunch of math and creating charts and all this kind of thing. That's probably something that's gonna require like more of a specialist approach because you know we're now dealing with sort of custom data stores and custom business logic that is more complex and more critical than than many office tasks tend to be.
SPEAKER_01Yeah. But I mean, I would assume that what we're talking about here is and what what I've seen coming in through the through when people come to talk to us from a professional capacity is that they have gone really far. They've built stuff that would have cost $100,000 two years ago. Two years ago. And there are yeah, they and they've they use make.com or N8N and a couple of GPTs or Claude or something like that uh with skills and a couple of mockdown files, and they have built something go a long way with a long one. Two years ago would have three years ago would have cost hundreds of thousands of dollars to build. And then the only reason they come to us anymore is like we I want to take this to scale, I want to like wrap this, I want to I've prototyped this whole process. I know what all these pieces are supposed to do. I've basically almost written all the pipelines and all the software myself because I am the expert. Yeah, I am the lawyer, I'm the banker, I'm the consultant, I'm whatever it is, and I've strung all these little agent things together. Now, you, dear software developer, the only reason I need you in the picture is to take what I've prototyped and put it into a way that I can measure, put it into a server where it's nice and secure and no one's gonna hack into it. Maybe take it out of these tools, like take it out of the make.com and put it into something that I can own and I can have intellectual property around. Yeah. And that's that's kind of like so we're like we we we're kind of like falling into that like lost mile kind of bit of delivery.
SPEAKER_04Where some right.
SPEAKER_01And I and there are some part, there are some parts of this that are more complicated. Like if your agent needs to go and interact with some deep data set, like you've got a I mean, this is awfully specific, but let's say you had a bunch of medical research data, like the guy we talked to today. Yeah, yeah, yeah, yeah, yeah. You have a bunch of medical research data that's highly specialized and is not well organized. You probably still need data analysts and statisticians and stuff to like get that data into a nice, clean place, and you need to put it into a repository, and then you need to have your agent be able to go and fetch that in a safe way and probably in a HIPAA-compliant way. You probably need experts to help you with that. But you can you don't have to you can prototype that on your own by getting some fake data that looks like that, or like you know, dumping it into an agent or into a spreadsheet and getting it to read from the spreadsheet. And then once you've got like prototype, you bring it to somebody like us, and then we'll take it to market. Right.
SPEAKER_04Then you can just kind of go that last mile. Yeah. Yeah. I mean, I I think that like another example of when you probably want to do that would be if you're sitting on top of a production data set that is yours and you like your database, um, you even if it's not your database, but like critical order data, banking data, any any kind of critical business data that like if something happened to it, you would be real screwed. Yeah. For those kinds of things, right? Like you might want to make like a real-time replica of that data, and your agent works against the replica so that if it does something, it's contained, right? And so these are things that require kind of more sophisticated infrastructure setup. Keeping a real-time replica of a database is not something that is just anybody can just do. That requires some technical expertise. But you don't have to do that until you get to that point. To your point, you could go a long way, you could prove this out on a spreadsheet for a long time and honestly operate on a spreadsheet for a long time as kind of your data store, as it were. And only when you're ready to kind of scale that up, then can you say, okay, let's like do this for real, which will end up happening because once you see how powerful it is, you're like, we gotta hook this up to the real sauce.
SPEAKER_01Okay, I got a I gotta one lost idea then to run past you, which is and then I I don't know if this is this is just getting to the end of the day. But we talked about we talked about tasks. We talked about them tasks being like the atomic uh elements of a job, right? And then we talked about how agents can handle tasks, string them together can handle jobs, and become turn them into workflows. And then something you said that that keeps sticking in my head is that you're not replacing that person and you only end up needing the software engineer when all these things become complicated and you send them out into the real world to face clients. Let me run this past to you. I think if you are replacing uh if you're using agents to replace internal tasks or internal jobs, you can almost entirely do that on your own with stuff off the shelf. The time when you turn into a software engineer uh is when that agent or set of agents or agentic workflow, whatever it is, is going to start uh interacting with clients and the general public.
SPEAKER_04With clients, so yeah, and or data.
SPEAKER_01Yeah. Maybe. So so so like, hey, I am going to augment, replace, or add assurances around a customer-facing person, a client-facing person, that's when you probably want um, or you're using getting close to real data, that's when that thing needs to be, you know, speak to a soft speak to a professional then.
SPEAKER_04Yeah, just to make sure that like, yeah, yeah. You you don't need it falling over in front of people. Yeah, like you can fall over in front of your staff.
SPEAKER_01You prototype your robotic um junior associate consultant and um internally, and you prototype him, and then when it's ready to actually go talk to real customers, that's when you and it's gonna get paid and someone's gonna pay that bot. Right. Then that bot needs to be built by somebody who knows what they're doing. Yeah. I don't know. Only you think about that.
SPEAKER_04But yeah, I mean I think that is uh I think that is that in 2026, I think that is true. Or if you're just too busy. Or if you're too busy or for other lots of reasons. You don't you're you don't want your job to become managing all the workflows and the technology and the agents let somebody else do that. Strangely enjoys it. Cool.
unknownYeah.
SPEAKER_04So that's it. When you're yeah, write your write your job descriptions, hire your agent, start thinking about your org chart as a blend of personnel types. Personnel's not the right word.
SPEAKER_01And go and go and go and try some of the tools out today. Go to make.com or n8n and open up a GPT and see if you can get them to read a Google Doc or Word doc and spit it back into the LLM and spit something back. And once you've done that, you will see the power of the stuff.
SPEAKER_04Absolutely.
SPEAKER_01All right, buddy. Well, it's nice to see you. Good conversation. I hope people learned a little bit from this and uh hope they're gonna stop listening and they're gonna go sign up for a make.com or N8N accounts. Hopefully so.
SPEAKER_04Enjoy it. And we need to like probably get them on a referral program or something because we have a lot.
SPEAKER_01I was gonna say there's there should be like a link below, but there's not. Yeah, they're right. I'm sure we can get it. Let's work out how to get that link below. All right, I'll see you next week. Thanks for seeing everyone. Bye.