How to Manage AI Agents Like Employees, with Brandon Metcalf, CEO and Founder of Asymbl | Ep. 435

0:00Welcome back to the Marking Millennials podcast recorded live at Salesforce Connections. I'm Daniel Murray and today I'm talking with Brandon Metcalf, founder of Assemble, a company with 106 human employees and 200 digital workers built from 2000 AI agents. Brandon's built a whole system for managing AI like real employees, job descriptions, coaching, even promotions, and it's added up to a projected$13 million in productivity this year alone. We get into what happens to human employees when AI takes on their tasks, why broad AI agents fail the way generalist employees do, and the one AI heel Brandon will die on.

0:47Let's get into it. Welcome to the Marketing Millennials, the no BS marketing podcast. I'm Daniel Murray and join me for unfiltered conversations with the brains behind marketing's coolest companies. The one request I tell our guests, stories or it didn't happen. Get ready to turn the f*** up. Okay, we're live with the Mark Millenials podcast at Connections. I'm here with Brandon Metcalf, CEO and founder of Assemble. Brandon, could you explain what Assemble does and how it's connected with Salesforce? Yeah. So thanks for having me. So Assemble, we're a workforce orchestration company, which is probably not a terminology you've heard before.

1:37Essentially, we do three things. We have recruiting technology to help you hire people. We have digital labor to help you onboard digital workers to help work with those people. And we have a consulting practice that helps you bring it all together so you can create this human digital hybrid workforce. We're connected to Salesforce. Our recruiting application is built on Salesforce. And then our consulting practice, we're a Salesforce SI, a top tier SI, implementing, customizing, deploying all the different clouds, agent force. We have a data foundations practice where Data360, Informatica, MuleSoft, and then we're really focused on agent fabric right now with all the stuff we're doing with agents and digital workers.

2:20So pretty deeply plugged into the ecosystem. I've been in the Salesforce ecosystem since 2004. So what made you bet on Salesforce as an ecosystem? Like you could have bet on a lot of different ecosystems. Yeah, and we're constantly corded to go to other ecosystems and expand, right? You know, the partnership, the relationship with Salesforce has been very fruitful for me over the 20-something years that I've been in it. This is, what, my fourth company that I've built in and around the ecosystem. For us, we keep coming back to Salesforce and keep building on Salesforce because of the platform itself, the trust, the security, reliability, the customer base, the innovation, things like Slack, Slackbot, now Headless360.

3:10It's just where we feel the most comfortable and the most secure for our customers that they can actually get the results that we're promising that they can get. Could you walk me through that human and digital piece that you were talking about? Because I think that's interesting because a lot of people are scared of like what is the human going to do in the situation what's they are going to do but like in your space recruiting what is like agents taking care of and what humans do so the story of how we got to where we're at right is and a couple years ago we were invited to participate at what at the time was einstein copilot which became agent force um to to build what is now our digital recruiters or recruiting agents on uh on agent force to work with our recruiting product that's on salesforce so we're starting to get into that and be in this beta program where agent force was really being created and then at the same time we're just launching the company trying to grow trying to scale i started using ai i started using all the gpts and all that to figure out like how can i move faster how can we accomplish more started building a lot out in that arena and started to really unlock the potential of what we could use it for so then got my leadership team plugged in they all started using it and then there was this moment where I was looking at this I'm like what I'm trying to get this AI to deliver for us in a business context is the same thing that I look for employees to deliver for us like there's a job I wanted to do.

4:44I know if I coach it, it gets better. I know if I stop coaching it, kind of goes sideways, which is very similar to employees, right? If I really manage it, and then if I put clearly defined job scope, motivations for success we talk about for that job, if I do all of those things, this AI, this agent, if you will, turns into an employee. So we started thinking about it that way. And we really started to rationalize, well, we want to scale the company we've grown like i don't know 53 54 x over the since we launched in in 2023 so quick growth about 150 160 human employees today we've got about 200 digital workers today we've got about 2 000 agents that make up those digital workers if i can explain that madness so what that is if you want um but we really focused on what are the roles what are the jobs what's the motivation for success should that role be a human role should it be a digital role, should it be a hybrid?

5:42And I can explain some of those examples. And we started to generate this value for us, this actual ROI that we can measure. So last year is about a$5 million productivity impact for us internally. This year we're forecasting it's probably going to be a$13 million impact for us, real numbers that we can quantify. So when we started to figure this out, and we started, of course, applying it to our recruiting products, and then we started seeing customers have similar needs, it led to us creating the digital labor practice, which is by far and our fastest growing practice that we have. And then we really were like, wait, let's take a step back.

6:16If companies are trying to use AI as employees, what happens to the employees? What happens to people? And what we saw with our team is even the entry-level employees become managers. So we tell the story of a digital worker named Theodore Frank. It's our SDR agent. so Teddy when he's doing a good job and Theodore when he's been a bit naughty and we do that because we get a little laugh right so everyone buys into it humanizes it but we're also setting the stage that Teddy is going to have bad days and he's going to have to be managed and coached and there's discipline around that so we had a human SDR named Mitch and Mitch's part of Mitch's job was when we were rolling out Teddy manage Teddy figure out how to make Teddy better and And what happened is Teddy got better, but Mitch got a lot better, which was already great when he started.

7:09But he eventually got promoted to an AE. And his first month as an AE, he closed like 11 deals. Because of the context and the understanding of what Teddy needed to do to be good at his job, Mitch was able to pick up all of that and apply it to his sales process. So we see humans evolving into managers of this because they need to understand the outcome that it's delivering. and now we're seeing the success that our customers are having in a similar way that we're having so it's a lot of fun. I also think your space is like one of the spaces ahead because you know how to write a job description you know like how to like coach managers and do performance reviews and all that stuff and that's pretty much what you have to do with AI is like set a scope of like this is what you need to do um here is like here's your here's like a like a type of performance review so it's it's it's i think recruiting's and and i've seen like recruiting and your space a little bit anti-ai too which is crazy like you like you're like people who are in recruiting and hr should actually or could be are like set up to be some of the best users of ai because they know how to do a job description they know how to coach they know how to manage like so why How are you dealing with that tension?

8:32It's evolving. I think the mindset is changing. So when we look at this hybrid workforce, human or digital, it's job descriptions, it's job design, but it's also organizational design of what roles do you need? Why do you need them? What's the outcome you're trying to deliver? How does this role impact that role? Which is HR's bread and butter of how do we org design this? The technical nuances as they're getting more normalized and simpler to like conceptualize and think about is making it easier for non-technical folks to really think about how they can use these types of workers to do their jobs.

9:09But I think recruiting itself is radically changing. Like now, well, resume, like we were, we wrote out a product called talent intelligence, which is our candidates really brain of like, how do you find the right candidate for the right job? Traditionally, you do that by taking someone's resume and matching it to a job description. And I did recruiting early in my career. And I remember when I was on the desk, I used to love terrible resumes because no other recruiter would touch that resume. And this is a while ago. so I would wait and get these terrible resumes like the person didn't know how to write about what they're doing their successes their well any of that stuff and I would have conversations with these people and I would pull out of them what they've done who they are and all of that I did a lot of recruiting for accounting and finance and like those types of roles and you know not always the most dynamic type of personalities when you think of accountants so pulling that out I could then use that to actually help find them jobs get them placed and do all that now with ai writing resumes like it it creates a similar type of problem where the candidate do they do they actually have this experience how do you know it's real how do you know it's validated so with our talent intelligence product we actually go and we say what about the rest of the relationship what about the phone calls what about the interviews what about the jobs that they've gone through a process?

10:30How far did they get? So we can start to put together this profile of a person, if you will, that actually shows more than just what is listed on the resume to match them to the job. And that's going over really well in the recruiting space because these recruiters are like, this is what we've always wanted. We've always wanted to see the rest of this information about the candidate so we can then figure out who we want to talk to. So I think not only is the mindset changing in the recruiting space that you can use AI to do this stuff. I think the technology is rapidly innovating that gives them real purpose to use it.

11:03So it'll be fun to see in the next year, I would say a few years, but in the next year, six months to a year, how much this radically evolves. Since it's a marketing podcast, I want to get your point of view from a marketing take. So a marketer is trying to hire employees versus AI. Like, how should they think about like setting up their org to like, and how should they do the hiring process with AI and not AI? So like, how should they think about hiring a human versus AI and like the sourcing process and like working with AI to do that in the background to make sure they find the best human or the best set up the best AI?

11:43Really, really great question, especially as you're framing sourcing for candidates in a marketing context, because that's exactly what it is. And, you know, we have a recruiting applications business, but we also help with non-recruiting processes as well. And then again, internally, we use our own thing. So tomorrow here, I have a session where we actually talk about our go-to-market process. So we're about to roll out a digital agent on our website named Fenn. So Fenn will engage with people on the website, get them the information they need, direct them, do all of those things, and try to qualify to see if they're a potential fit for what we do and vice versa.

12:22Once that lead is created by this digital agent, we have a second agent that is then the Salesforce prospecting agent, where it will go and it will then enrich everything we can find about that prospect so that then the third step of the process, Teddy, the digital SDR that I was talking about, can take that lead and Teddy can start marketing to it and start to try to really qualify them into ICP and score, grade them, our likelihoods to buy, all of that. And then Teddy, once that's all qualified, will try to schedule the meeting with the actual sales rep, the human, and hand that off. And then from there, Jordan, the human SDR that we have on the loop, gets to manage all of Teddy's work and success and failures and all of that to teach him and coach him and make him better.

13:14So our whole process is web agent, prospecting agent, SDR agent, then to a real AE or human AE. And then we have another digital worker, Casey, who's a marketing guardian, if you will. So Casey's whole job is to review all the contents being created to make sure it's within brand guidelines and to provide a report basically to the humans managing this process to say, this is on track, this is off track, this is how you should coach any other agent. agent throughout this process to speed up our ability to manage this process. So it's a cohesive workflow of four different types of digital workers or agents, one human in the loop to manage that whole funnel.

13:57Yeah, I've seen some of like, I think you made a good point earlier about that an agent is like a human and it needs to be coached and like consistently. And I think the biggest problem, and I want you to answer this for people who are building agents, is people are just spinning up agents out of control. They have 200 agents, but they don't know what the agents are doing. Agents are talking to agents, but there's no human in the loop. There's no manager. So the agents are wild talking to. So what do you recommend? How many agents to a human? um where is the human in the loop of it because i think that's like there's going to be one side of the picture where people are going to say agents suck because they don't know how to manage it and there's other people like you who are setting it up where there's actual process yeah um yeah i just want you to yeah and the way you described it i would think agents would suck um so like like we i always make fun of this is the answer is not let's just slap an agent on it and have the agent do the work.

15:10Like that won't work. You hit on it earlier about job design, workflow design. And I think this is why it clicked so well with us from coming from the recruiting space of like, what's the job to be done? What's the motivation for success? Because if you don't have a plan, you're going to have zombie agents, meaning someone spends all this time creating an agent. And then a week later, they forget about it and no one touches it. And it's just turned into a zombie or you have drift. so we rolled out early days a marketing assistant that i built and got it exactly how i wanted it so i could ask it questions about who we were what we did and all that mind you our positioning messages changed a lot over the years but it was really really good i could use it i got what i needed super quick deployed that to the team and i stopped managing it and all of a sudden the team was creating all this marketing content i'm like what is this like how are like are you not using the agent?

16:05And the answer is like, no, we've been using the agent. This is what it's producing. The agent started to figure out what they wanted and it started to like drift and create stuff that it thought was good. That was terrible. And that's really one of the steps that we learned early on. We have to manage, we have to coach, we have to have to discipline, we have to have the guard rails to control it. So we don't have that. So I think having a real thought of what do you want this to do? What's the job? And everyone talks about jobs to be done and tasks. That's part of it. What's really the outcome that you want it to do?

16:40And how do you make it successful doing that? And who's going to manage it? And if you start with that, and we advise this with clients a lot, usually clients will come to us with the most grandiose, complicated use case for a digital worker or an agent. And we scale that back. We're like, let's focus on one real meaningful outcome. So if we can get it to deliver and you can manage it and coach it, then we can expand from there. Because what it does, it teaches you how to do it, but it also gives you what we call a validation of belief. That it can function like an employee. That you can manage it like an employee.

17:14And then you can get the same outcomes that you'd get from an employee depending on what the role is. Every single time we've done that, it's expanded to 10, 15, 20 different agents or digital workers doing the work. And it's been a big success. but anytime we don't go through that process the risk of drift the risk of zombie agents is huge and i this is in marketing too but i think recruiting is like one of like the most deeply human experiences because like you're dealing with people making really it's their livelihood yeah their livelihood um so like how do you manage the difference between like optimizing for speed and efficiency, but it also doesn't damage that experience of like they're going through like actually one of the biggest decisions of the other.

18:00That's the whole point, right? It's the whole point of the design process of what are you trying to accomplish? Where at every company is different, like you have Amazon hiring a bunch of warehouse workers. Like what's that hiring process versus hiring like an executive? Very different type of interviews, questions, relationships, trust, all of that. So it's really focusing on what's the success criteria for whatever role you're hiring for and the job that you're setting up. Recruiting is deeply personal. And what makes really good recruiters effective is their ability to connect at a human level with the candidate and with the hiring manager to deeply understand personality, what they're really looking for, what they're not saying, what's between the lines, all of that.

18:47So we talk a lot about judgment, but we also talk a lot about taste. Humans have taste, meaning a sense of quality and all of that type, not like a taste layer. Yeah, so they have a taste, they have judgment. We're in the process, do you need to apply that? People talk about high value and low value tasks, and I think that's a terrible way of looking at it. Because like any of our digital workers, the tasks that they're doing, I don't see any of it as low value. If they're completing the job, that's high value. But it's where do you need those human specific traits in the process so that you don't lose sight of at the end of the day, businesses, at least in the world today, are humans working with humans.

19:30So there's a piece of that has to stay human. I get asked a lot about what's going to happen to the junior workforce. Where does that go? and I refer back to like the Mitch story of what happens is you start to hire people to teach them what a really good outcome is for a particular type of role and you then teach them how do they manage the tools or the AI that they have in order to execute those roles and the human management piece I wrote a piece about the what I call the talent paradox I think that is going to become a skill set and a rarity that's going to make recruiting challenging to find people because how do you build that knowledge how do you build that muscle if you will if your company is not already embracing ai so for me recruiting is always going to be human but there's a lot of repetitive process steps that you don't need those human-centric skills to do yeah i mean we talked about it before like sourcing is like one of the like finding the town that is a deeply manual process like especially if you're like sort of going through linkedin profiles and sourcing does this match like our job description that's a deeply a deeply but it like you said also like that taste there for humans like there's something about a human gut in marketing in recruiting that just knows like this like based on their experience based on their their experience based on like logic based on data that they have they can make a really good decision that ai is going to probably tell you to the logical ways here yeah you got to be careful with you know just like the ai can write a resume ai can write a job description so like if if a company is just using AI to write all of their job description, does the job description really reflect what they're actually trying to hire?

21:31Or is it just the AI's interpretation of what they're trying to hire? And then the matching piece is special because there's been a lot of conversations in the past about bias and AI making assumptions and AI repeating patterns. So really being diligent about how do you make sure it's not applying bias? How do you make sure it's surfacing all the candidates and helping you identify, you know, what you should be looking at without eliminating candidates based off of some criteria that it's judging them on based off of previous hires and stuff like that. So there's a balance with how you do it. But a lot of it is gathering data, processing data, visualizing it.

22:17So like if you think about applications, so we hired 100 people in 100 days in the beginning of last year. we had 17 ,000 applications. We had one human recruiter. There's no way a human alone could go through 17 ,000 applications. So we applied our technology, we applied Agent Force, we applied all of that to actually help us say, okay, who's just completely a not fit? Like this person lives in a different country or this person, you know, applied for an engineering role and they are a salesperson, like all of that stuff to slim it down. And then how do we go deeper and how do we go deeper to speed up our process to find the right candidate pool.

22:55So there's definitely processes and tricks to doing it, but there's the balance of at the end of the day, like you said, this is one of the most important decisions a person will make, what job. And you want to make sure you're looking at all candidates and really finding the best candidate based off of bias, not based off of a resume that's written, but really based off of what experience they have and what experience you're looking for. I also want you to go into, I know you use Slack as like a communication tool to work with digital workers and agents. So can you tell us how you set that up to make like a quality interactions between everybody in the company and the agents to make sure it works officially?

23:37Yeah. I mean, so for us, Slack is a far more than a communication tool. It's our operating system. So when we first started using Slack a few years ago for the company, it was communications, but then it was, if you look at our company, We have people in the United States, fully remote, like I was saying. We have people in India in an office and also remote. So you have different time zones. Then you also have just different types of jobs they're doing. So you have sales and marketing and all of that, that they're actively in Salesforce, customer success actively in Salesforce. That's the system they use.

24:11Engineers are using other systems like Jira or whatever to write tickets and write code and all this stuff. So they're on different platforms. but we have similar processes that need to happen across. So expense approvals, new hire requests, like the 100 people in 100 days, we actually set up in Slack the requisition process to request a rec, get budget approval, do all of that. That gets signed off on, then it would automatically create and recruiter suite our recruiting product, the rec, so the recruiter could start working on it. The recruiter can then use Slack to communicate and ask questions and clarify and do all of those types of things, regardless of time zone.

24:48and then go through the hiring process and being able to see what's happening in the hiring process and Slack or inside of the Salesforce UI, get to an offer. We use Slack for approval processes for the offer to communicate and all of that and then get to hire and onboard the employee. They start, they're set up in Slack. We have Slack hubs and all of that with videos and links and all of that. So it's one centralized place that everyone can see. And then now with Slack AI and more importantly with Slack bot, Because we've been doing all of that in Slack and it's now connected via MCP, so our Gmail and our calendar and all this other stuff, all of that context is there.

25:27So it makes finding information super easy, but Slackbot can execute on things for us as well. So it really is not only our operating system, it's basically our command center, not only for the humans, but we have all of our digital workers or most of our digital workers accessible there. So Casey, the marketing agent I was talking about, you can go into Slack and ask Casey to write you a LinkedIn post. Or you can ask Casey, is it okay if I say this or that? So the entire team, regardless of where in the world, regardless of what time, has a central place to communicate with this marketing specialist that gives them approved content that we can trust.

26:05So Slack, in this interface, in this way of working, makes us really operate with the company 10 to 20 times our size. I think the one point that I didn't think about that much because I don't work with people like eight hours, nine hours is the time saving with time zones. Like how much, because you, I have worked in companies like that and you, you wait not eight, seven hours. Yeah, you stuck for seven hours to get approval. You stuck for seven hours to get a request in and you stuck for seven hours to ask a simple question. And they ask the question, then you're done with your work day. So you had to go do it.

26:40Just the impact on the business. That's like two, three days of waiting to get a simple answer. My first company, a company called Talent Rover, we had nine offices in eight countries, literally around the world. And I founded that company in 2011. It sold it in 2018. This was a huge struggle for us then. I learned deeply business slows down dramatically, not because people aren't working hard, but because the working hours they're working just doesn't overlap. So how do you normalize that? So Slackbot helps us do that a lot. yeah i i i never thought of that but i think that's such a clutch thing to do and i was wanted to ask you um since you've been using ai for a few years now how many agents are reporting to you and like what what a part of your job description have you off-bordered to agents and what are you doing now as your job like what has changed since you're using yeah i mean and and you know, I've interchanged the word digital worker and agent.

27:39So let me describe that a little bit. So for us, digital workers are actual, basically collections of agents grouped together, orchestrated together to complete a specific job to be done that we can define motivation for that we can coach. So for example, one of my digital workers is one named Bradley. Bradley is a digital intelligence analyst. So Bradley, anytime I want, I can say, tell me how the company is doing. who go into 15 different systems, process information, analyze information, and come back with a 10 to 15 page briefing for me on everything from sales performance, specific deal performance, individual rep performance, what's going on with client projects, what's going on customer success, what's going on with cashflow, where are we at versus budget, all that stuff.

28:23So he comes back with this report and then I can ask him questions and I can go into like, well, I'm, I'm concerned about cash this week. Like, what do I need to be aware of? And have this conversation. So Bradley's been trained in all of this. Now, Bradley originally had about 50 different agents under him. And what we learned is when Bradley was spinning up his subagents, they weren't always that great for specific parts of the process. So one area that he struggled with was around sales and understanding really what's going on with the rep, what's going on with the specific deal. So we learned that Bradley needed a different digital worker and we created Stanley.

29:04And Stanley managed this whole sales process. Stanley has a fine job, motivation for success for that job. I now manage Stanley. I manage Bradley. And then Stanley processes that. So when Bradley, when I asked Bradley for his report, he'll go and gather everything and he'll talk to Stanley. And Stanley has to produce Stanley stuff to come back to Bradley and Bradley produces that back for me. So, you know, we have about 200 digital workers. So the collection of these agents together throughout the company, some are way simpler than that. Some are like a single agent that has a specific job. Some are super complicated like Bradley.

29:38We have about 2000 agents across the company. We have a chief digital labor officer. Have you ever heard that title before? His name is Shiv. His whole job is deployment and management of our fleet of digital workers. Shiv's background, he was a former founder and CEO of a Salesforce consulting firm. So I didn't hire a chief AI officer that was just technically focused. I hired a builder. I hired someone that knew how to create workforces, who also has a technical background. And he's deployed all of these digital workers. So my job, and everyone in my company's job, We've all changed. We've all evolved.

30:15All of us manage now, junior level to me. My role now is much more being present than I could ever be for my team, holding them accountable, driving motivation, being in front of clients, doing things like this, and meeting with our investors, really driving the business forward and eliminating a lot of the administrative burden that I have. I still have an executive assistant, Barbara. She's been with me forever. Her role's been able to evolve. You can take on more operational things. So at the end of the day, we've all been able to do more than we could ever do before. And that's really the mission of my company is how do we help other companies do more?

30:54How do we make their humans elevated with the use of digital teammates? So how many do I directly manage? I mean, Bradley and Stanley are probably only two that I directly manage. Slackbot I use all the time, but the team uses quite a bit. Yeah, and those two agents alone probably would have taken you before asking. Just that field briefing? Yeah, would take 10 different humans and meetings. In weeks. In weeks, yeah. And then the data's stale. The data's out of date by the time I get it. I need to make decisions right now. So Bradley can take anywhere from 30 minutes to like an hour and a half to run, depending on how much I'm tasking with.

31:33But back to like the conversation around time zone and business stopping. Me, I need real information to make decisions. sometimes on the spur of a moment waiting weeks for our team to go and crunch this data to give me the answer is not practical so the speed of which this data can be compressed and given to me makes us move just so much faster i have two more questions um one how did you like reverse engineer that that bradley had like a sales like the the problem like that failure yeah failure so one of our core values. We have a couple of different core values that are related to this. Felfast, and then also bias towards action.