We interview the founder behind Megan, an AI HR agent • The Register
Interview Mega HR, a Florida-based human resources startup, today launched an AI agent service called Megan that the biz claims can automate most recruiting and hiring tasks while improving communication with job applicants.
While we’re disappointed the firm didn’t go with the spelling used in the title of the 2022 horror film M3GAN, about an AI doll that becomes self-aware with predictable results, we nonetheless welcomed the opportunity to question founder and CEO Darren Bounds about the consequences of adding AI to the already fraught recruitment process.
There’s widespread sentiment that the hiring process is broken. Job interviews are a nightmare, it’s said. Folks have spent months applying for hundreds and hundreds roles without success. Ghosting – the unexplained cessation of communication – is common, as are ghost jobs – positions posted that won’t be filled. Meanwhile, 74 percent of employers in one recent study claim they can’t find the desired talent. In light of that, perhaps the question is whether AI can make things any worse?
Bounds in an interview with The Register said he previously founded recruiting platform Breezy, which was sold in 2019 to Learning Group Technologies.
Bounds: “I was going to leave HR tech permanently after that company was acquired and then COVID happened and LLMs happened. Really, the whole kind of hiring space was disrupted and changed in a lot of ways, much more rapidly than it had in prior decades. It seemed like an interesting opportunity to start rethinking what things could look like.
“About two years ago, I started working with agentics [or software agents], testing the boundaries of what was possible. It became pretty clear very early on that there was an opportunity and that opportunity was changing every month because of all of the innovation happening on platforms like OpenAI and their ability to reason. [It also became clear] that [AI] agents are going to become a very important part of the SaaS landscape very rapidly over the course of the next five years.
People have been apprehensive about using the technology
“So about a year ago, we started working on Mega HR. And the vision behind Mega was that, okay, agents are definitely going to play an important role in hiring and marketing and sales and all of the above. But all of these companies that are currently using traditional SaaS products are going to have varying levels of acceptance of AI in their workflows.
“For some people, AI has been part of hiring for a very long time now in different forms, pre-LLM. But people have been apprehensive about using [the technology].”
The goal, Bounds explained, was to build an agent capable of performing as much of the hiring workflow as possible while also allowing companies to adopt the technology gradually, in accordance with their level of comfort.
“So we started work on the Mega platform itself, which is an end-to-end hiring platform, the same as you would find with any other existing applicant tracking system, like Ashby or Workable. But with the mindset that we wanted agents to play a role in all of the capabilities that we built for that product.
“So today we have a system in place where every feature that we add to our product is automatically learned by Megan and incorporated into her tool belt so that she can, if you’re interested as a user, manage that part of the process for you.
“And that includes everything from crafting and advertising jobs, publishing them, editing them, to managing vetting candidates in the early and later stages of the hiring process, certainly scheduling all of that, which are really table stakes.
She has an understanding of your business, your team that’s involved in hiring
“But also, she has an understanding of your business, your team that’s involved in hiring, the candidates who are in motion, and can play a role in interacting with any of them using the tools that we’ve built.
“So what does that mean? You may have, say you have a variety of candidates who are in a particular stage. They’re in the interview stage. and you need to gather some information from them and update your notes before a meeting that the team is going to be on to interview them.
“You can ask Megan, exactly like you would a person sitting next to you as an assistant or another member of your team, to gather the information from [the job candidates] and update the notes so that the team has it before the meeting.
“She’ll reach out to them. If it’s an email, she’ll email them. She can text them. She’s completely multimodal. And she’ll capture that information, update you throughout the process. If she runs into problems, she’ll let you know. Otherwise, the notes will be updated and you’ll go into your meeting with everything that you need.
“So we’ve built Megan with a very different philosophy than I think a lot of the current, at least the publicly visible agentic solutions have been built. And that is to guardrail her around certain things that we never really want her to delve into in terms of conversation or capability.
“But within the capability and scope of what we want her to be able to do inside of the hiring process, we let her use the tooling that we built for her however she feels is necessary in order to accomplish the task that’s before her.
“And we’ve spent a lot of time over the last six to nine months, really refining her ability to reason through those processes and deliver what’s usually a favorable outcome.
“So it’s actually been fascinating. We use her for hiring. Other companies are using her for hiring in our pilot phase. But it’s been really, really fascinating as a builder and as a science fiction fan to interact with her and see what’s possible today through agentics. And we’re just on the tip of the iceberg in terms of this capability, where we’re going to be in 12 or 24 months from now.”
The Register: What’s your pitch to companies? Is it essentially that you’re saving employees X amount of time a week?
Bounds: “So, the value prop is definitely in time savings for team members.
“What we look to primarily is how much of the process that you’re involved with, from posting a job to hiring a candidate, can you outsource to Megan?
“That’s ultimately what we look at … and, depending on your hiring process, that can obviously vary tremendously. [Megan], in total candor, is what I would consider to be the MVP state of capability right now. There are many, many more things that I would like to add, and we will add, over the course of the next 12 months. But within the scope of things that she’s able to do today, and that’s kind of how we’ve chosen the companies that are using [Megan] today, [the metric is] how much of that process can she take on from a human?”
The Register: What’s it like to deal with Megan for job applicants? Do you have any feedback from them? Among those applying for jobs, there’s some frustration with the level of automation and the lack of transparency in the process.
Bounds: “Even before the current state of LLMs today, that’s been a huge problem. Because applicant tracking systems for, I would say, since 2008 or 2009, have been incorporating machine learning into their processes and disqualifying candidates based on what, before the existence of LLMs, were very, very rudimentary decisions.
“But now, especially in this job market today, ghosting candidates is a huge, huge problem.
“A big part of my product research has been done through Reddit and watching people talk about the horror stories that they’re going through, the hoops that they have to jump through to not get a job, and how much time is wasted.
Some companies just may want to go radio silent
“One of the goals right from day one around Megan was that we would provide a very consistent reliable path to updates and information about where people are in the hiring process. Now that doesn’t mean to say that we’re force-feeding every company that uses us to abide by that. Some companies just may want to go radio silent.
“But the default path and recommendation from Megan is that each stage in the hiring process has SLAs tied to them, goals for moving a candidate through those stages. And when those SLAs are violated or missed, they trigger Megan to automatically communicate statuses to candidates.
“And not only that, she’s proactive in that sense and that she’ll keep the candidate updated throughout the process. But the candidate has a direct line of sight to Megan and can communicate and ask her questions which she will respond to and go to the team and the humans that are involved with it and share and share with them their queries and go back and forth between them to try to manage that relationship so that if there are questions, they can be answered.
“But without question, there will be updates when [candidates] move through different phases and they’re either in scope or out of scope for the hiring process.”
The Register: In the process of developing Megan, did you run into any problems, failures, or errors? LLMs are known at times to go off the rails.
Bounds: “Those are the challenging parts, right? A lot of AI products are very shallow, smoke and mirrors, one-shot AI solutions. Some people get a little bit further beyond that, but very rarely. And that’s why we’re so excited and proud of what we’ve done.
“We’re building software to try to address deficiencies in reasoning and hallucinations in LLMs. And at the same time, though, the platforms building the LLMs are also doing that. Their work has been, without question, very, very valuable in the process. The level of reasoning available today compared to 12 months ago from an agentic perspective is night and day.
“I don’t even necessarily know how to evaluate it. I don’t have a metric. But we’re in a completely different place [today]. At the same time it’s required us, based on where things were six months ago or 12 months ago, to build many many layers to help the agents.
There are dozens of micro agents that work together to create what is Megan … she’s guardrailed on a micro scale, but not at a macro level
“Megan is not an agent. She is a concert of agents. There are dozens of micro agents that work together to create what is Megan. They’re all focused on different subtasks that they’re very good at, and they work together. And within each one of those, there are technologies that we’ve built or strategies that we’ve implemented that allow her to reflect and validate that what she thinks or what another agent thinks to be true or isn’t true and then take action on that, all the way through the chain to ultimately revealing something or taking action with a candidate or revealing something to the user.
“But those were much more significant problems nine to 12 months ago. There are still problems today, but ones that we’ve guardrailed around in a pretty interesting way. We use the term guardrailing kind of hesitantly because one of the things we love about Megan is that we don’t tell her how to do anything.
“We don’t tell her how to email a candidate and update a candidate record after they’ve gotten feedback from a candidate. She has to figure that out. We provide her with a bunch of Legos and she can put those Legos together in whatever way she wants to solve a particular problem. So she’s guardrailed on a micro scale, but not at a macro level.”
The Register: What’s the onboarding experience like? Is there software to install? And how is the service priced?
Bounds: “It’s all web-based. So you go to our website. You’ll be onboarded and given access to the product. You can configure it or you can be yourself or use one of our customer success managers. And then you basically manage that yourself all through the cloud.
“The pricing model is based on the size of the organization. And then on the agentic side of things, it’s based on the number of candidates that Megan has brought into the in-progress state. So any candidates that you’re hiring for in a job that are disqualified automatically, there’s no cost associated with those. But any candidates that she is managing for you, we have a cost associated with that.”
The Register: What about the compute cost for running Megan?
Bounds: “I would say there’s a lot of risk to us right now in that process because this type of capability is brand new, especially with an agent that is as robust as Megan. For every interaction with Megan, there’s a cost [for our company] and how often you interact with her will vary substantially between organizations and jobs and teams that are involved.”
Bounds explained that Megan gets injected into products at a macro level so users can interact with her through Slack or sales CRM software or in other ways. A decision was made to adopt a simple pricing model with static costs rather than account for every LLM token.
“We’re assigning a fixed cost to make it simple and understandable by the customers when we talk about this because it’s a new technology. I would say there’s going to be a lot of uncertainty about what the costs could look like and by doing it in this way we feel like it’s as probably as close to value-based pricing as we can probably implement today without having a lot more knowledge about how actual interactions are going to look.”
For a company of 10 to 30 employees, Bounds suggested the cost might run between $200 and $500 per month depending upon the sophistication of the company’s hiring process.
Bounds: “Just recently, over the last two months, we started using Megan for our own internal hiring. And this kind of mirrors what we’ve been hearing from the [20 or so] companies [that have been testing Megan].
“Initially, you start out interacting with Megan very delicately. You ask her certain questions. You configure her to perform certain operations, like screening candidates or setting up interviews.
“But eventually you get to this point where you have a problem that you explore, asking Megan if she can do [a particular task]. And she does it. And you’re like, ‘Oh, I need to rethink how I’m using this software,’ Megan starts to become less of a feature that you are using and you start to think about her more as a partner in your process. And once you go through that inflection point, it really changes how you start really testing the boundaries of what she can do. And that’s when it starts to get really exciting.”
The Register: Has the scope of what Megan can do become broad enough that companies are actually replacing human workers with AI agents? Are they hiring less staff because they can accomplish more with fewer people?
Bounds: “I think there’s no question that if you asked me that six to 12 months from now, the companies using [Megan] would say yes to that. But because we’ve only had companies using the product for six months, it’s too early to say that, especially since it’s January and budget seasons have just started.
There’s a lot of benefits, but there’s going to be a lot of change
“What it does do, I think, is that it scales your solution in a way that would affect your decision to add capacity to your hiring team, or at least add capacity in the same manner that you would. It really frees your team to do less of the repetitive stuff and start focusing more on actually establishing relationships and understanding the people that are in your hiring process, which I think is a really good thing.
“This technology is completely disruptive, not just in the context of what we’re doing. It is evolving very quickly, and it is getting very, very good at replicating the behaviors that you would expect of a mid-level employee in a variety of ways.
“And so my answer to your question is ‘no, we haven’t observed [human roles being replaced with AI agents] with existing customers today.’ But I have zero doubt that the answer will be ‘yes’ for any customer using and leveraging Megan 12 to 24 months from now.
“And that’s going to be happening across all verticals. There’s going to be a lot of uncertainty around what these technologies mean for our society. There’s a lot of benefits, but there’s going to be a lot of change.” ®
