If you’ve not listened to Round Up before, it’s a short review of the episodes that I’ve published in the last month to make sure you don’t miss out on the valuable insights that my guests are sharing.
This month Round Up returns to its live format, and this is a recording of my live conversation with Jim Miller VP of People & Talent at Ashby about four of the episodes published in September 2026
Episodes featured in this Round Up:
Ep 823: How AI Is Disrupting Entry Level Jobs
Ep 824: Is AI Creating Talent Debt?
Ep 825: The Real Risks Of Candidate Fraud
Ep 827: Agent To Agent Recruiting
Follow this podcast on Apple Podcasts.
Follow this podcast on Spotify.
Transcript:
Matt Alder: Hi everyone, and welcome to this month’s Recruiting Future Roundup Live. Brilliant to be live again and talking to everyone today. If you’ve not watched Roundup before, it’s a summary of some of the best episodes that have appeared on the Recruiting Future podcast over the last month or so. Once more, I’m delighted to be joined by an excellent co-host. Jim, welcome to Roundup.
Jim Miller: Thank you very much. Thanks for having me, Matt.
Matt Alder: A pleasure. Could we kick off with you introducing yourself and telling everyone a bit about what you do?
Jim Miller: Yeah, I’m Jim Miller. I lead the People and Talent function for Ashby. Ashby’s an ATS, and I’m pretty certain most of the audience will have heard of us. I live in the US, despite the accent. I’ve been in America for nearly 11 years now. I moved over with Google, I spent 15 years with them, and I’ve been at Ashby for the last three and a half years.
Matt Alder: Fantastic. So we are looking at four episodes that have been on the podcast in the last month. It’s really interesting, actually, because we’ve covered a few diverse subjects over the last few weeks, and I think there are definitely some interesting things to talk about here. Looking forward to the discussion.
So we’ll start off with episode 823, which is called How AI Is Disrupting Entry Level Jobs, and it’s an interview with Reuben Ogbonna, who is the co-founder of the Marcy Lab School. This interview came about because I was talking at the ERE conference earlier in the year, and Marcy Lab School were there as an exhibitor. I had a really interesting conversation with them, and we thought it would be great to do a podcast. What they do is they’re an alternative education institution that teaches people tech skills, particularly AI skills, and helps them get that first job in tech. So it’s really set up as an alternative to going to college. In the interview, we were talking about how entry level hiring and entry level work is really changing in the tech sector. Jim, what did you think of this one? I know it’s something close to your heart, as it were.
Jim Miller: It is, and it was really interesting because obviously I look a lot at what’s going on in the world of AI, but it felt really analogous to me to about 20 years ago. I joined Google as a tech recruiter to build out the office in London, and there weren’t any engineers at this point. What I realised while watching our recruiting programs teams is they were doing exactly this kind of work. They were going into universities to teach the professors what technologies we were using, because those professors wanted all of their students to come and work at Google. It was interesting that we were, in effect, doing exactly what Reuben was describing there.
What I found back then is that it was the students who took control. Students were going out and doing their own open source coding, building their own projects, doing their own hobbyist work, publishing this stuff, and that was what we were using to determine whether the person was strong enough in this space. It was in effect giving them the professional experience in coding, getting code reviews and everything else on what they were producing, and it was a fantastic way for them to showcase their skills. So they were circumnavigating what the education system was forcing on them, and I think that’s probably what’s going to happen now. The younger folks are going to be picking up these AI skills, they’re going to be building their own portfolios of technology that they’ve produced, showcasing apps or whatever else it happens to be, and that’s going to be their leverage point. So the starting point that institutions like Reuben’s are giving them to get them up and running, and then enabling them to do this work, I think will avoid the gap that we’re going to see in the entry level jobs market.
Matt Alder: Yeah, I think that’s really interesting. What struck me was the speed at which they could adapt their curriculum and the way that they worked to keep up with the market. On the part about entry level jobs changing, there were a few things that we talked about here. The feeling was that the entry level work wasn’t disappearing, but the bar was getting higher. New starters were being asked to do things that mid-level people might have done in the past. Is that something that you’re seeing? Is the nature of those jobs actually changing with the technology?
Jim Miller: Yes and no. There’s still the very, very highest echelon of students who are being taught by the people who taught the real trailblazers in the AI space. Those folks are being picked up for entry-level roles. It’s a build versus buy kind of scenario. Companies could wait two years and then have to spend millions and millions of dollars to hire these individuals. So that’s still happening.
On the other side, in the work that I’m seeing, I’m seeing engineers move back to writing the code themselves and using the AI as more assistive technology for the code reviews, testing, pressure testing the way that they’ve solved the problem, and research on what the best solution might be. I think we’ve seen with certain companies having to lay off people because they’ve overspent on AI tokens that the cost of the AI doing all of the work actually outstrips the cost of the human doing some of this, and it’s natural economics. I think that’s the swing that we’re seeing. We’ll start to see more humans being brought into the space, so that it’s not all being done by AI, because someone’s got to pay the bill for all of these data centres that are being built, right?
Matt Alder: Yeah, I think that’s interesting. It’s something we’re actually going to cover a little bit deeper when we talk about the next episode. Before we move on, there was something else that really stood out for me here that is a recurrent theme. Reuben was talking about the skills that people have to learn, and actually judgement, collaboration and communication have become even more important in this post-AI world that we live in. How can people learn that? How can people get that? Is it something that they can be taught in the education system? Is it something they have to learn on the job? Has AI shifted the dynamics here? What’s your view on that judgement piece?
Jim Miller: I don’t think so, and I’ll touch on this when we talk about the next episode as well. We’re an asynchronous remote company, and we still find ways for individuals to meet and interact, virtually or in person. So that’s key. But I think it comes all the way back to the opportunities that you have, for team sports or any other element. Whatever you can do to throw yourself into social situations is going to help your development naturally, so that when you then get to this world of work, the collaboration between the humans is going to remain critical, in my opinion. The creativity piece of two humans coming together and creating something that wouldn’t have been produced by either of them is still the powerful point.
Matt Alder: Yeah, I couldn’t agree with you more on that one. It’s probably a good point to leave this one and move on to the next episode, which was actually the next episode of the podcast. So episode 824. It’s an interview with Matt Campbell, who is one of the managing directors at Alvarez & Marsal. Interestingly, it’s called Is AI Creating Talent Debt? We covered a few topics in this podcast, but the central piece to that title was what you were saying about big tech companies laying lots of people off because AI is changing everything and the way that they work, and then having to hire those people back again or rethink it. Lots of companies are copying that model and being burnt by it, and that’s the talent debt part of it. Talk us through that a little bit.
Jim Miller: I think some of it’s mislabelling. I think there was a big reset of wages based on the wage inflation we saw in 2021 and early 2022, and companies are just looking at it and going, well, we can cut these folks and rehire, in effect, the same positions to reset there, and very much using AI as the excuse.
But I think the underlying piece that we just touched on is the collaboration. There was a reason why Google had long lunch tables, and it was to throw people together to have conversations. People who wouldn’t know each other normally, wouldn’t have a chat normally, bringing two different mindsets, two different sets of experience together to create something new. Certainly in the early days, some of Google’s biggest products came out of that kind of interaction, over a game of pool in the office, a game of foosball, over the long lunch table. You can see it in apps like Donut in Slack that bring groups of employees together on a random basis. I have two of them later on today, and I love those moments where you can get together.
So I think that’s the big pain point, because when you strip out huge swathes of an organisation in the guise of a layoff, you’re not necessarily picking who’s going. It’s just the job that disappears rather than the humans, and you are going to lose so much there. Those gaps will take a while to show themselves, by which point those people have gone off and had new opportunities and probably driven careers that would make them not want to go back into that organisation. So it’s an odd state of affairs, but I think companies are going to learn from it very quickly. What it will do is tighten up how they hire in the first place and make them much more intentional about getting the best out of the folks that they do bring in.
Matt Alder: Yeah, it came out really clearly in the conversation, because we were talking about a job being a bundle of tasks, but actually a role is the role that someone plays in the organisation’s social system, and that was a big part of it. That’s something that these kind of paper exercises just can’t see, and it’s such a critical part of this.
The other part of this conversation was really around the commercial value of AI and how people are implementing it, because that’s something that Matt works on the strategy for. There are two bits to this, really. The first one was around AI pilots, and how they mostly end inconclusively because people are using them just to test whether the technology works, not whether it actually solves a business problem. What’s your view on that? Obviously, you work for a software company as well. What’s your view on pilots that you do, or pilots that you see your customers do? Are people getting it right?
Jim Miller: No, they’re not, because the hypothesis is the important thing when you’re doing something like this, and the hypothesis is generally that we’re going to save money. What they don’t do is intentionally repurpose the time that will be saved by the technology. Matt touched on it a lot during the podcast. The analogy I use is combine harvesters. I grew up in the middle of Suffolk. The farmer buys a combine harvester for the first time, and he’s got this labour force that used to do the harvest. If he lets them sit on the side of the field drinking cider, he’s got zero ROI from the deployment of that technology in the three seconds that the combine takes to do a day’s work for an entire team of humans. That’s the point. You have to isolate it. So if you put those workers to work building a mill to mill the grain, so you go deeper into the supply chain, or diversifying your business and growing other crops that are more labour intensive, you’ve got an ROI.
It’s Parkinson’s law of productivity. Work fills the time available to you. So if the human is given two hours back per week, they’re going to do tasks that previously weren’t a priority. Ergo, you’ve got no ROI. Flip it around the other way. If you’ve got 20 workers and they’re each saving two hours per week, if you can isolate one person and take them out of that group, that’s the 40-hour work week. You repurpose that person somewhere else. That pilot enables those 40 hours to be saved, and now you have a force multiplier going elsewhere. That’s true ROI. So how you frame the pilot and how you isolate the expected time saved is the only way to prove or disprove your hypothesis.
Matt Alder: It’s so important, because I made a list of the kind of phrases that I get back from people when I ask, if AI is going to save all this time for your team, what are they going to do that’s different? You tend to get very generic things like they’re going to do the important work, or they’re going to do what they always should have been doing. When you dig a bit deeper, there’s just nothing behind there. So it’s so critical, specifically in TA, to really think about where we add value to the business, how we can redeploy people and really, as you say, get the ROI from this.
Jim Miller: For me, we’ve got a whole suite of AI functionality, and each piece is working in series together to give a compound impact on the productivity of our team members. It means I probably have a smaller team than I would have done for a company growing at our pace five years ago. It means that I can reinvest that time in quality. Our operating principle number five is that we’re frustratingly patient about hiring, which to me is a mandate for quality. So I’m able to refocus what our team does and how we hire. For example, we build cohorts of candidates before anyone gets put into an interview. It’s the equivalent of the UK civil service model, but I’ve deployed it in a B2B SaaS company, to try and create a shortlist at the end of the process of multiple candidates who are above the bar, to create choice. Choice drives the highest quality. I’m able to do that because we can spend the time building the pool at the beginning and then work incredibly quickly using the assistive technology and the tool to get to that stage. Ergo, we end up with stronger results than we were expecting from the workforce that we hire. Those are the two sides of talent density. We’re not trying to hire the smallest workforce. We’re trying to hire the workforce we planned for, and then get better than expected results. That’s the outcome from our deployment of AI technology earlier on in the recruitment process. We’ve thought deeply and intentionally about that.
Matt Alder: Yeah, and what I like about that is that you have basically used the technology to have a completely different process to give you those great results, rather than using the technology to make your old process quicker. I think that’s such a great example of what’s possible.
Let’s move on to the next one. Slight change of topic, but something equally important. So, episode 825: The Real Risks Of Candidate Fraud. It’s an interview with Lauren Furey, who is the principal product manager at Proof, who make technology to help protect people from candidate fraud. The first thing to say about this is that up until a few weeks ago, I think a lot of people just lumped a lot of things together, so fraud, cheating and using AI in the recruitment process all seemed to end up in the same bucket. What I think is really important about this conversation is that it very much pulls out the fraudulent, criminal, dangerous activity that we are seeing in the recruitment process. A few months ago, people would have thought of this as something that wasn’t affecting many companies, something that was a bit of a one-off, but I’ve seen the volume turn up on this discussion. What we mean by fraud here is basically criminal and state actors targeting companies to try and get jobs to get inside their systems, for whatever purposes. So organised criminal gangs, the North Korean government. It’s happening, and more and more I’m finding companies who have had this problem. Obviously, this is core in terms of what you do as well. How big a problem is this, and are people taking it seriously enough?
Jim Miller: It’s huge, and they are. Just for ourselves, we did an analysis last quarter, and I think 16% of applications for software engineering roles we deemed fraudulent using our own system, which has a different way of working to Proof’s. It was flagging 16% of them as potential to high potential fraud, and specifically around the person not being the person they said they were. So again, I’m using that tight definition that you’ve just used.
Matt Alder: Yeah, exactly. The other thing that struck me about this is how ill-equipped a lot of TA teams are to spot this. You see lots of people talking about it, asking how do we avoid it, and lots of almost home-baked remedies, like getting someone to put their hand in front of their face. But these are quite sophisticated attacks, and you need the proper tools to combat them, don’t you?
Jim Miller: You do. Am I able to tell you a little bit about how Ashby does it?
Matt Alder: Yeah, of course.
Jim Miller: Proof’s model is very much proving who the person is through traditional ID verification. Ashby’s model, which has been live for well over a year now, is an analysis of all of the inputs coming through, all of the signal inputs you get when you apply, and it takes every single element line by line. There’s no stack ranking or anything else there to identify potential issues. I won’t go into the nuance of it because I don’t want people to start to reverse engineer it or try to attack it using the information I’m giving. But when the person is reviewing the resume, all of these signals are displayed to the user, so they can then make their decision.
A lot of the conversation here was around how IT departments are, in effect, leaving the recruiter to take the action themselves. This gives the recruiter all of the information that the IT department would have themselves, and then you can build your own risk tolerance rubric as to who you let through the system and so on. So it’s an early intervention to give signal, and then if you do want to take it further, you can go on and do a secondary ID check after the fact. I think that’s the really interesting piece here. You’ve got some solutions that are ID verification first. That creates a barrier to entry, which is potentially in conflict with the whole concept of recruiting, which is to hire someone downstream. Because if the person who is great chooses not to apply, you’ve got a missed opportunity. So that’s the needle that we were trying to thread with this non-invasive initial approach, to give the signal directly to the user, and then they can follow up with the validation. But they’ve already got the candidate in hand and can start building the relationship.
Matt Alder: Yeah, I think that makes a lot of sense. There’s obviously a balance here between candidate experience and protecting the business, and also no one likes to be accused of being a fraud if they’re not actually a fraud. So there’s obviously going to be a debate here, a balance here, different ways of working, all of those kinds of things. But it’s obviously really important that TA has tools available to do this, and also presumably really important that they’re working with other areas of the business as well, so there is a joined-up approach to this. Does TA just get forgotten about and left out, or do you think companies are identifying that as a route into their security?
Jim Miller: Yeah, absolutely. I mean, we have a hotline to security. We can flag anything at any time. But it’s interesting, this is happening across functions. I’m hearing about people’s products becoming a vector through the sales process, to get access to the product and use the product in nefarious ways, across the B2B SaaS landscape. So there are attacks coming into pretty much every department, which is elevating security as a function, but it’s also elevating the conduits into each of those departments to truly understand what their processes are, to then nip all of this in the bud. That to me is the really interesting scale element, because ID fraud in employment is not new.
Matt Alder: No, not at all.
Jim Miller: We saw it with human to human, where you had employees who were selling their own services to do interviews because they knew how to pass them, and then somebody else was coming on board. Or even the case of twins, where one person does the interview and the other person comes to do the job. It was fascinating stuff, and this was still when you were human to human in an office.
Matt Alder: Yeah, exactly. So it just shows how easy it is to do in some respects. I think also when we were talking about the future here, this is only going to get worse as AI really accelerates the way that people attack companies.
Jim Miller: Absolutely. I was chatting to somebody the other day on a trip to London, and he was in security, and he was talking about their patch cadence, and about how they used to patch every year, and now they’re already down to patching every day.
Matt Alder: Wow.
Jim Miller: That’s in a two-year time span. I don’t know anything about this stuff, so I couldn’t comment on the technical elements, but to me that was just, wow, what a magnitude of change being caused by the capabilities of the nefarious actors. He was telling me it was actually the good actors who were flagging the system problems so that they could then fix them faster. So there are two sides to it.
Matt Alder: No, absolutely. I think the overriding message from this is that people should listen to this episode, and if you are leading a TA function and you do not have tools of some kind to combat this, then you need to look at that urgently, because these are attacks on an industrial scale against lots of businesses.
Jim Miller: There is also the labour cost, because you’re still going to process every single one of those applications. If you’re spending 25% more time this year than you were last year on application review, that’s a problem that you’ve got to solve.
Matt Alder: Yeah, absolutely. It’s going to become an even bigger discussion point as we move forward, I’m sure of that.
So let’s move on to the final episode that we’re going to talk about, which is 827: Agent To Agent Recruiting, an interview with Matt Wilson, who is the co-founder and CEO of Jack & Jill. I actually recorded this in their office, and I think it was the day after they had done a big raise, so they are a really interesting organisation. I want to talk about this in a broader sense, because agent to agent recruiting, and I’ll explain what I mean by that in a second, is something that I think is the future. I’ve recently got back from RecFest USA, where I did a presentation about agent to agent recruiting and how I think it’s developing. The idea here is that the candidate has an AI agent that helps them look for jobs, and the employer has an AI agent that helps them look for candidates. That’s a very simple way of describing it, and I think it’s the direction everything’s moving in. What Jack & Jill have done is they’ve almost built a closed system of this, so it’s almost like the next generation of job board, if you like. They have one agent that talks to candidates, so everyone who registers gets a 20-minute conversation with AI to get some more information, and they have another agent that then helps employers find people. So it’s fascinating, and I think it’s a microcosm of where the industry is going. What do you think?
Jim Miller: Funnily enough, I actually met Matt two weeks ago in London, on camera, and he was interviewing me for their website, so that’s kind of full circle. Far too many Matts this week as well. I think it’s really interesting because, going back to basics, it’s hard enough to get a job, let alone a job that you’re not qualified for, and that to me is the crux of the situation here. Where I’m seeing the AI influence on the candidate side, I’m seeing the creation of collateral that supports that individual for a job, where that collateral isn’t grounded in truth. That’s a real problem for us, because in order to determine whether that person has that capability, we need either very advanced AI to do it on the employer side, or a human. The joke, even as long ago as two years, was that you post a job and one minute later you’ve got 5,000 applications where your system says they’re all qualified. We’re going to need more humans.
I think there’s a third element to this agent to agent piece. You’ve got the input from the candidate side, you’ve got the input from the matching engine agent, and then I think there’s a third piece, which is agnostic, which is the judgement part. That’s going to be the really interesting piece. You’ve got AI interviewers who are pre-programmed to do the interview that a human would do, to avoid scale issues and so on, which is fantastic, and they’re able to dive deep and give objective feedback about that individual. That’s the third piece that I think will have to come into play here. I don’t think that the employer’s first agent will be combined with the second agent. I think they will deliberately separate those two out to sort the wheat from the chaff, to use another agricultural analogy, and then drive the recruitment process from that point. I think that’s the change that we’ll have to see. On the other side, the democratisation of jobs, the ability for people to find new jobs, to be out there in the marketplace and be given the opportunities, can only be a positive thing. We’ve seen it with the remote work revolution, and I think this discovery and matching piece is just fantastic.
Matt Alder: Yeah, I think there are a few things around this. One of the things that I’ve noticed is that candidates obviously have access to a lot of tools now, many of which are helping to drive this noise and signal problem that we have. I think I described it as looking for a needle in a haystack that is made completely out of needles. So I think that from a TA perspective, a lot of people are looking at this the wrong way. They’re looking at this candidate technology as just people grifting, exploiting candidates, spam. But in amongst all of that, there are some really interesting companies building candidate technology, and I think a lot of the innovation in the sector is going to come from that, because at the moment a lot of companies are adopting AI processes because they’re trying to deal with the things that the candidates are causing. So I think that’s interesting, and I think that anything, as you say, that gives people the ability to find more opportunities, and also to look at opportunities that they wouldn’t have considered but are actually really qualified for, is fantastic.
Then on the employer side, I agree with you. I think what this does is enhance those human conversations. It makes them happen quicker. Hopefully it avoids the situations where people sit down and say, tell me about your resume, talk me through your last few jobs, and all that sort of stuff. So I think there is a real positive thing here. I also think that it’s something that TA needs to be prepared for, because there are so many ways that approaches like this could go horribly wrong as well, in terms of who owns the data, who owns the persuasion. I was reading something yesterday that someone had written about personal AI agents and Facebook’s Muse, which was frankly terrifying when it came to psychological manipulation and things like that. So there’s a lot going on here, but I think it’s important for TA teams to realise that this is the potential direction that we’re going in, and they need to be ready for that.
Jim Miller: They do. I’m old enough to remember being allowed to take a calculator into maths tests and exams at school, and it’s a tool. So it’s about bringing the tool in and embracing the fact that the tool can be used in the work, and you want to hire the people that are using the tool in the right way. That’s key. So bringing the tool into the process, the tool in general being AI and a multitude of different agents. The critical piece, I think, is that TA teams need to separate out what the base criteria of the human experience is. Going back to earlier in the conversation, the experience that’s been built through collaboration and through the journey through the career and so on, isolating that piece out, and then being very, very transparent as to what they require the human to provide, and then what they will allow the AI to be part of. We’ve tried to play with that a little bit with application forms recently, using subtext and nested questions and a whole bunch of other things to really showcase the humans that are thinking very deliberately about their application and the experience they’ve had, versus those that have just deployed agents to throw as much mud at the wall as possible, so that we can avoid having to spend a lot of human time on the needle in the haystack full of needles.
Matt Alder: Absolutely. So we’re pretty much out of time, because we’ve gone through them all in a lot of detail. I want to just finish, though, because you’ve mentioned this a few times as we’ve gone through the conversation, with your own recruiting process, because you mentioned you copy bits from the civil service. Tell us how it works and why it’s different.
Jim Miller: So we have evergreen roles, the roles that we’re constantly hiring for, and that’s all down to a calculation of time to fill and how quickly we need the next one. That sits to one side. For every single one-off role, we will post the job with a job description. There’s a narrative style that gives an enormous amount of information, including why you should not apply for the job. Deep, deep transparency, and I’ve got data around how that transparency impacts things downstream. Every job has a closing date, and we give the instructions to the candidates to take their time over the application, because the application will be the tiebreaker, not the CV.
We then let them know that we will review 100% of the applications after the closing date. No need to message anybody, nothing at all. Everybody gets a fair crack of the whip, and there’s a statistical model behind that, because if you only review the first 10% of the applicant pool, the likelihood that the best candidate, as subjective as that statement is, is in the first 10% is very unlikely. Likewise, the best candidate from any particular demographic, however you want to slice and dice demographics, is unlikely to be in the first 10%. So we’re capturing the entire pool here. We source before the job goes live, so we have a pool of candidates that we like the look of from sourcing. They get the outreach on the day the job goes live, so they see that the outreach has happened on the day, we call out the closing date, and we’re asking them to apply. So to the point around the huge amount of information, the spam going out and all the noise, hopefully this cuts through that. Referrals and internal candidates all go into the same pool.
Then we stop, and we have the time set aside to review that entire pool we’ve created. We use data to inform how long the job is posted, by the way, and that time is set aside beforehand. We use the technology to go through. The first pass is all the people who aren’t qualified. The second pass is long list to shortlist, still objectively. Then the hiring manager is involved. They have a number, and they’re allowed to move N forward. So they might have 3x in the pool, and they’ve now got the difficult job of choosing who goes forward to that next round. The moment they do, we release the time that’s been held for the interviews. Everyone gets the interview scheduling automation at exactly the same moment, and they have free open calendar space to fill. So now that cohort has been built, and they will move through the process at roughly the same pace, because they’re all interviewing within 24 hours of the release of the interview times.
What that enables is, first of all, hiring managers never come back and say, I’d like to see a few more. It gives us choice at the end of the process and keeps everybody together. I think it’s much more objective and fair, and we’re never stack ranking the candidates. Everybody gets past each stage and moves through the process, and the only time you would make a choice is after the hiring bar has been met, we’ve had our debrief, and we’ve decided we’ve got three candidates who are above the bar. Which one does the hiring manager and the departmental head want to hire from that group?
Matt Alder: Makes sense. Fantastic. Jim, thank you so much for joining me. It’s been brilliant talking to you. Thanks to everyone who is watching this, or if you’re listening to it on the podcast. Lots of great content coming up on Recruiting Future in the next few weeks, on some really big topics, so if you’re not following us wherever you listen to your podcasts, please make sure that you do. Finally, Roundup Live will be back again this time next month. So thanks very much, and speak to you all soon.
