As technology accelerates, the future of work is evolving from simple AI assistants to fully autonomous agents. In this follow-up to our previous session on AI with member John Arnott, CEO of C1M, we’ll explore how workflows, agents, skills, and tools interconnect across every layer of an organization. Join us to gain a clear framework for integrating modern AI-driven solutions, optimizing workflows, and strategically deploying autonomous agents where they add the most value. Whether you’re looking to enhance productivity, streamline operations, or future-proof your firm, this session will provide practical insights and actionable strategies to keep you ahead of the curve.
TRANSCRIPT
Greg Alexander: Hey, everybody! This is Greg Alexander. You’re listening to the ProServ Podcast. This is a show for founders of boutique professional services firms who would like to make more money, make scaling easier, and someday achieve an exit. On today’s show, we’re going to continue a series in the broad category of artificial intelligence. We’re going to focus on how, currently, at the end of February 2025, we seem to be moving from AI assistants—things like ChatGPT, Claude, etc.—to fully autonomous agents. This is a real opportunity for members of our community at Collective 54. We have a great guest with us to discuss this, John Arnott. That name is probably familiar to you because he participated in this series previously and introduced us to his AI maturity model, which I’ve been using and has been super helpful. Today, we’re going one step further and building upon that. John, great to see you again. Thanks for being here. For those who might not know you, could you introduce yourself, please?
John Arnott: Sure. Thanks, Greg, appreciate it. I’m John Arnott. I am the CEO of C1M. We’re an organization dedicated to helping businesses grow and transform using artificial intelligence. We often start with their marketing and work through their back-office operations from there.
Greg Alexander: Alright, perfect! John and I spoke before we got on the call today, and he walked me through this thing called the agent stack, which has four layers to it. I like to think of it as layers of a cake because I love a good chocolate cake. We’ve got the process layer, the agent layer, the skills layer, and the tools layer. John, if you’re alright with it, we’ll use that as a framework for our call today. First, tell everybody what an agent is.
John Arnott: Sure thing, Greg. We’re in this era where we’re transitioning from assistants—where you go to a tool like ChatGPT, Claude, or Copilot, ask it something, and it gives you something back or analyzes something for you in an interactive chat experience—to agents. Agents have agency, meaning they have the authority and capabilities to do something on behalf of a person. An agent is often autonomous, meaning while it could be triggered by some activity, it operates on its own timeframe and is responsible for accomplishing a goal. It’s responsible for output. When I think about an agent, I think about its resume and KPIs because it is responsible for doing something within a certain context. So that’s an agent—it’s a component that gets something done on behalf of another agent or a person.
Greg Alexander: Okay, fantastic. I’m going to give the audience an example of how we at Collective 54 are using agents. We run our financials using QuickBooks, as many of our members do, and our corporate credit card is an American Express card. We now have an automated data entry agent, so when receipts come in from American Express, they automatically get categorized correctly. We used to have a partner in the Philippines do that for us, but we no longer need that, as the agent is now handling it. That’s just one example to bring this to life. Now that we know what an agent is, let’s go step by step through your layers of the cake. I’ll start with the first one, the process layer, because you educated me on how foundational it is. If you try to implement agents without well-documented business processes, you can get yourself in trouble. Walk us through that one.
John Arnott: Yeah, that’s great, and your example was excellent. This first part is the workflow or business processes. You want to think about agents as a tailwind to help get something done quicker or more effectively. If you add that tailwind to a bad process that agent is not going to make things better; in fact, it’ll make things worse. For those organizations that do business process reengineering and those that advise on business processes, what they do is more important than ever. We want to ensure organizations have solid, fundamental processes, both departmentally and organizationally, that can be augmented by agents supporting the activities within those processes.
Greg Alexander: Yup, okay, I’m going to give everybody another example. Again, I’m going to stick with QuickBooks here because everybody knows that. We produce a cash flow forecast. Our friend Ken Yeager, inside of Collective 54, has taught us about the importance of a 13-week rolling cash flow forecast. Previously, when we did that, it was manual, error-prone, and, to be honest with you, a source of friction between me, the owner, and my employees who had to do that. So we now use the cash flow forecasting agent within QuickBooks, and after some tweaking and fixing, it produces a cash flow forecast in 5 minutes, and it’s 100% accurate every time. So there’s another example of how we put an agent together. Now, that wouldn’t have worked unless we had a forecasting process. Again, we use Ken’s 13-week process, but you have to have that because the agent can do whatever you tell it to do, but you have to tell it what to do. All right, so that’s the process layer. Let’s go to the agent layer, which was step 2 in your tool that you sent me.
John Arnott: Yeah, so within a process, we often have people and technology working together well before these AI conversations. That’s how we’ve always done things. Then there are typically handoffs between people and often software within the process. Where does an agent fit in that? A person has a job, and within that job, they do different activities, have different responsibilities, and perform various tasks. An agent is responsible for some of those tasks and activities that support that job. It fits within the process framework because, in many cases, a person will hand off to an agent, which may hand off to a series of agents, which may hand back to a person. That agent layer is the orchestration layer. It’s the one that coordinates a group of skills together to get something accomplished, fitting within the overall business process that it’s supporting.
Greg Alexander: Yup, okay, another example for the audience. We’re coming up on tax season, and of course, I want to get as many deductions as I possibly can. Within QuickBooks, they have this thing called smart recommendations. They have the tax code embedded into it, and they looked at all of my expenses, determining what I could deduct and what I could not. I had recently moved to a new home, and I was wondering how, because I work from home, I could deduct my home office. It calculated it for me by determining the square footage of my office as a percentage of the square footage of my entire home. It was amazing to see it do that. That’s an example of an agent doing something—not just doing something faster than I can do it on my own, but doing it better than I can do it on my own because it had the tax code in the tool. I’m not going to read the IRS tax code, but it did, and it showed me ways to save money, which was fantastic.
Greg Alexander: All right, number 3 of 4 is the skills layer. I think this is where we get tripped up. So, John, tell us what the skills layer is and what we should be looking for here.
John Arnott: Yeah, you actually gave a really good example of it with the tax code. In that case, you have an agent, and one of its skills is its knowledge of the tax code. When you think about an agent, there are many components. One of them is a knowledge base, and that knowledge base is used for its purposes. Whatever it does, the skill is the capability to do something with that knowledge base. That skill can often be thought of as a smaller process. When we do a task or have something in a role that we do in a job, these are often mini processes with 3 or 4 steps. That’s a skill—putting things together to support what the agent’s doing. That skill uses its knowledge base, long-term memory, short-term memory, reasoning ability, and communication ability, either with other agents or people. That’s what we mean by a skill, and it’s abstracted away from the next layer, which we’re about to talk about: tools.
Greg Alexander: So let’s jump into that—the tools layer. The mistake I think people make, and I’ve made this mistake myself when I’m less disciplined, is lumping skills and tools together. They need to be separated. Tell us what the tools layer is and why tools and skills need to be bifurcated.
John Arnott: Yeah, so the tools layer consists of the systems and resources that the agent actually accesses and utilizes. For example, I mentioned that an agent can communicate. The tool might be that it communicates in Teams, Slack, or email. The agent may do something in QuickBooks. The tool, in this case, is a component in QuickBooks, such as “get account” or “run report.” The skill is to analyze the tax code, and that’s independent of QuickBooks because I can analyze the tax code as a skill of any underlying financial accounting system. QuickBooks would be the tool that supports the skill of analyzing the tax in this case. That’s how they’re related—it’s the actual tool that it’s using to do something within a skill set.
Greg Alexander: Yeah, very good distinction. I mentioned taxes because it’s that time of year, and I was talking about how I am personally trying to improve my effective tax rate using agents. But we also at this time have Collective 54 members sending us notes like, “Hey, I need an invoice for XYZ.” Because membership dues in Collective 54 are a deductible business expense, they need to validate it with an invoice. Intuit has this thing called Intuit Assist, and Intuit is the parent of QuickBooks. You just literally prompt the system, and out pops the invoice. I mean, it’s absolutely incredible what this thing is capable of doing. Intuit Assist is the tool, and the skill is producing the invoice on behalf of the members. It’s just a great way to think about agents.
Greg Alexander: So let me recap. The agent stack is the process layer, followed by the agent layer, followed by the skills layer, and then the tools layer. I want to briefly tie this back to the previous episode, John, that you were on, which was the AI maturity model. If you think about those four components of the AI agent stack, how does one reconcile that with the maturity model? For example, if I’m through steps one and two, where am I maturity-wise?
John Arnott: Great, excellent question. Moving from assistance to agents means interacting with different layers of the maturity model. To reiterate quickly, our maturity model level one is interacting with documents. Level two is back-end systems. Level three is data. If I think about an agentic system, an agent has a knowledge base, either shared with many agents or its own. This is typically a level one type of project or interaction. However, that agent, within its skill set, will use different tools. A tool could be a CRM system, an ERP system, or a proprietary system. If the tool interacts with an API, it might be level two. If it interacts at the direct data layer, it could be a level three tool. So, I may have an agent with level one access to documents in its knowledge base, but it uses skills with tools that are level two or level three, depending on the organization.
Greg Alexander: Very interesting. That was a good way of thinking about it. I’m going to throw you a curveball here. Maybe this is my last question, and we can end on this one. When I think about agents, the other things I’ve read suggest thinking about them as a digital workforce or digital workers. For whatever reason, that clicked with me and helped me understand what an agent is. Then, because I’m an old guy and have spent many years in corporate America, I think about the org chart. Where do these digital workers fit on the org chart? Are they at the top, middle, or bottom of the pyramid?
John Arnott: They actually fill all different areas of the pyramid. In fact, I have an org chart for both a sales organization and a marketing organization that I was recently working on. It has the roles and the components that the agent does. We color-code them because there are certain activities the agent does with human oversight. A human who previously did a repetitive task may now oversee 10 agents performing different repetitive tasks. This changes the org chart. The seats occupied by a person now require a very strong level of communication. They must be great communicators, not just great doers, because they interact with the agent, providing feedback that goes into the agent’s long-term memory so it can learn. The org chart becomes a bit bigger and denser because agents fill roles that may have been combined into a single human role before.
Greg Alexander: Yeah, you know, I think that emphasis on being a good communicator is such a great piece of advice. As I look across the Collective 54 community, which I have the privilege of doing because I interact with all the members, I see different levels of adoption. Some are pioneers, way out on the bleeding edge, while others are on the opposite end, saying this will never work. Then there’s a whole bunch of people in the middle. Those maximizing this opportunity the most are naturally fantastic communicators. The level of frustration is reduced because they get it right the first or second time. They’re able to clearly tell the agent what they want versus partially explaining, which forces the agent to figure it out on its own. That doesn’t seem to work so well. Would you agree with that?
John Arnott: Absolutely. I often say that the next superpower is one being able to interact face-to-face with people and two, being great communicators. Because that’s the world we’re moving into, where the better communicator you are, the better results you get from people and agents.
Greg Alexander: Yeah, I agree with you. All right. Well, hey, John, if anybody wants to pick your brain on this, what do they do? Should they go to your LinkedIn profile? Should they go to the Member Portal? What would you like them to do?
John Arnott: Well, I always encourage members to go to the Member Portal and message me there. Of course, if you want to connect with me on LinkedIn, I’m John R. Arnott II, the second. And of course, you can always message me, too, at JA@c1mai.com.
Greg Alexander: Okay, fantastic. Well, listen, on behalf of the community, you’re such a great member. You’re always willing to contribute. I appreciate you being here today.
John Arnott: Yeah, it was great talking to you, Greg.
Greg Alexander: All right. So, a couple of calls to action for the rest of us. If you’re a member and you want to participate in John’s Q&A session, look for the meeting invitation that will come out for an upcoming Friday. If you’re not a member and you want to become one, go to collective54.com and fill out the application, and we’ll get in contact with you. But until next time, I wish you all the best of luck as you try to grow, scale, and someday exit your boutique professional services firm.
Note: This transcript was generated by Zoom.