If you're a TA leader trying to make sense of agentic AI in recruitment, the hard part isn't finding tools. It's working out which of them do what they claim.

That's the thread running through this episode of the TA Disruptors podcast. To close out the season, host Robert Newry sat down with Adam Gordon for a deliberately honest conversation about AI agents in hiring. Adam founded and scaled Candidate.ID (acquired by iCIMS in 2022), now runs Poetry HR, an AI workspace for recruiters, and co-hosts BrainFood Live with Hung Lee. He's known for calling the recruitment tech market as he sees it.

The backdrop: Randstad recently estimated that 1,200 new TA tech tools launched in the US in the last 12 months. Most will never be independently reviewed. So this conversation is about how a TA leader cuts through that noise, starting with the buzzword on every vendor's homepage.

What is agentic AI, in plain terms?

An AI agent is software that completes a task on its own once you've set the instructions. It can use a bit of initiative, but it's still doing a job you designed. It is not a colleague.

Adam's starting point is that the term is running ahead of understanding. In a room of 30 experienced resourcing leaders, he found only about 4 in 10 could confidently define agentic AI. That matters, because you can't evaluate the effectiveness of something you can't define.

It also means the boldest sales pitch collapses on contact. As Adam puts it, the narrative that an agent is "almost the same as an employee" is product marketing, designed to plant the idea that "I can get rid of Sarah and John if I have this AI agent." An agent completes tasks. It is not a person.

Why isn't most "agentic AI" actually agentic?

Because a lot of what carries the label doesn't do what the label implies. In Adam's view, nearly every tool in the market is really doing one of three things: writing up meeting notes, sourcing candidates, or assessment.

His take on sourcing is blunt: much of it looks exactly like the tools that existed in 2016, advanced Boolean search and keyword matching that scores a profile against a job description. "It's not really doing anything dissimilar to what was being done 10 years ago. It's just got AI written on it."

Two questions cut through it:

• Which of the three is this really doing, and what's genuinely new versus what we could already do a decade ago?
  

• What does it do today, not in a roadmap?

How do you tell hype from a tool that actually works?

Start with what it can do right now. Adam's own principle is to market only what a product can achieve today. A feature coming "in three months" is acceptable if the vendor says so plainly. A feature that's six months away but marketed as if it's here now is, in his words, a red flag, and the sign of an agent dressed in the emperor's new clothes.

He's candid about why this happens, and careful not to aim it at any one company. It traces back to the venture capital model: a fund expects most of its bets to fail and a few to return it many times over, so the pressure for rapid growth and hype runs from investors, to founders, to go-to-market teams. Applied to the 1,200 tools Randstad counted, Adam's rough maths is sobering: perhaps 200 will succeed and around 30 will do very well, which leaves a lot of tools being sold hard on stories rather than evidence.

His warning is worth quoting: "I can detect it quite quickly. When product marketing does not match product capability." The tell isn't the demo. It's the gap between the demo and what the tool actually does once it's yours.

Do AI agents replace recruiters?

No, and the pitch that they do misunderstands how agents behave. Agents break. They need monitoring, because a change in an AI model, a connected system, or a process can snap the whole workflow.

So the realistic shift isn't a team of 20 collapsing to one person plus software. Adam's picture is a team of maybe 15, plus two capabilities most TA functions don't have yet:

• An engineer to look after the technology.
  

• A talent product manager, a recruitment operations role, to design the workflow and keep improving it.
  

His line for buyers: you might get a team of 20 down to 15 with agents, "but I would need an engineer in my team, and I would need a talent product manager in my team." Buy the agent, and budget for the people who run it.

Why can't you automate all recruitment the same way?

Because recruitment isn't one job. Adam splits it into segments, agency versus in-house, across volume, experienced and executive hiring, and the room to automate changes with each.

Volume, in-house hiring has the most scope: he points to fully automated processes like Amazon's, where a candidate can meet their manager on day one, or a distribution model where day one on the job is effectively the assessment. Executive search has the least, because the recruiter is brokering a relationship on both sides. His summary is a useful test before you buy anything: "You couldn't just take Amazon's recruiting process and drop it into HSBC and expect that that's going to work."

There's a supply-and-demand layer too. Where talent is abundant you can ask candidates to do more; where it's scarce, you roll out the red carpet. It's the same reason a callback from earlier in the season holds up: make the process frictionless for the recruiter, but not effortless for the candidate. A little friction is how you see who's genuinely motivated.

Where is TA's value heading?

Towards the work that agents can't do. As teams get leaner, Adam is clear on where recruiters should point their skills: recruitment operations, candidate experience, and employer brand.

If your core skill today is interviewing or sourcing, that's the shift worth making now. Recruitment operations, the policies, processes, workflows and change management behind the tools, is where a lot of the future roles sit, and it's a capability TA has historically under-budgeted, borrowing it from IT instead.

How do you evaluate a tool if you're not an expert?

Lean on other people, and on the machines themselves. Adam's practical advice is to build a network of peers and ask what they actually use, not what they've been pitched. Then, he says, do the ironic thing: use ChatGPT to write your own diligence questions. Tell it "I don't know what I don't know" about, say, agentic sourcing tools, and ask what you should be asking to know whether a tool will help and whether it's ethically sound.

That discipline is a thread across this season. A couple of issues ago, Dr Charlie Eyre showed us how to test whether a hiring tool is valid, reliable and fair, and to ask for the technical manual before you buy. This is the same muscle, pointed at the newest buzzword: how to tell whether an "agent" is even an agent.

Key takeaways

• Define it before you buy it. An AI agent completes a task you've designed once it's briefed. It isn't an employee, and any pitch built on replacing named people is marketing, not reality.
  
• A lot of "agentic AI" isn't agentic. Most tools are doing one of three jobs, notes, sourcing, or assessment, and much of the sourcing is 2016 keyword-matching with a new label. Ask which of the three it really does.
  
• Judge it on today, not the roadmap. A feature six months out but sold as if it's here now is a red flag. Watch for marketing that outruns capability.
  
• Agents don't cut headcount for free. They break and need monitoring. Budget for an engineer and a talent product manager to run them.
  
• Recruitment isn't one job. You can automate volume, in-house hiring far more than executive search. Don't drop one segment's process onto another.
  
• Point your team where the value's going. Recruitment operations, candidate experience and employer brand are the durable skills as teams get leaner.
  
• Borrow expertise if you don't have it. Ask peers what they actually use, and use AI to generate the diligence questions you don't know to ask.
  

Listen now 👇

Adam and Robert cover far more than we could fit here, including whether recruitment is really broken, the ethics and bias problems in AI assessment, and why the best candidates now have to work harder to stand out.

Listen to or watch the full season 5 finale below.

If you're working out how to evaluate the tools in your hiring process, explore how Arctic Shores can help.

Transcript:

Robert Newry - Arctic Shores Co-Founder and Chief Explorer

Adam Gordon, Founder, Poetry HR

Robert Newry: Welcome to the TA Disruptors podcast. I'm Robert Newry, co-founder and Chief Explorer at Arctic Shores, the task-based psychometric company that helps organisations uncover potential and see more in people. Today we're wrapping up the series, and to close it out I wanted a proper debate. So I brought on someone who likes to say things as he sees them. Adam Gordon is well known to many of our listeners, especially those who follow BrainFood Live, which he co-hosts with Hung Lee. For those who don't know Adam, he's founded and scaled three recruitment technology businesses. The most recent before Poetry was Candidate.ID, a marketing automation platform for TA, which he built over five years and sold to iCIMS in 2022. He now runs Poetry, a workspace for recruiters that uses AI to save them around 25% of their daily time. He posts regularly on LinkedIn with often contrarian views on where recruitment is heading, and you'll get plenty of those today. We're going to cover a lot of ground: the explosion of AI tools for TA, some uncomfortable truths about what some vendors are actually selling us, and what "human in the loop" really means. Welcome to the podcast, Adam.

Adam Gordon: Robert, thank you very much for having me. I loved that introduction.

Robert Newry: Let's kick things off with the scale of the TA tech landscape. I listened to a Randstad webinar recently where they revealed that in the last 12 months, over 1,200 new TA tech applications have been launched. There's this huge rush for everyone to come up with a tool for every single part of the landscape, and it's getting really confusing. What's your take on the explosion of TA tech, and how might a TA leader make sense of it all?

Adam Gordon: The TA leaders who are going to thrive in this context are the ones who are really great at prioritising and filtering out noise. They set a plan and identify the technologies that will help them achieve it, as opposed to getting distracted every day by new opportunities from methods they've never heard of. TA isn't the biggest addressable market for software developers, finance and marketing are much bigger. But the reason we have so much noise is that everyone has been involved in recruitment. Everyone's been a candidate, lots of people have been hiring managers, and because most people have had a bad experience at some point, the common theme becomes "TA is broken, and I know how to fix it, because this happened to me and it must be happening to millions of others." So it's easier than ever to create technology products, and everyone thinks they know how to fix TA.

Robert Newry: Do you think recruitment is broken, or are people using that as a reason to launch something and create a fanfare about something that isn't as broken as they claim?

Adam Gordon: People are using their individual experience as market research, deciding we need something new, developing it, and putting it into the market, and often it's not the problem they thought it was. Recruitment is massively more complex than 99% of people realise. I'm not sure "broken" is the right word. Is it lagging other professional disciplines in terms of standards? Absolutely. A lot of the time TA teams operate very unproductively, people doing the same task over and over when they could do it once and share the workflow. Most people haven't gone through their own company's candidate experience. They don't realise that one-click apply opens the floodgates to people who aren't committed. There's a lot of common sense lacking, though there are pockets of brilliance, and it's improving every year.

Robert Newry: A lot of people come into recruitment because they see themselves as "people people", the skills were empathy, intuition, connecting with candidates. Now it seems to be much more about automation, technology, process and compliance. So it's not necessarily that recruitment is broken, it's that the things we've been doing are no longer appropriate. And there probably isn't a one-size-fits-all, is there?

Adam Gordon: I couldn't agree more. You couldn't take Amazon's recruiting process and drop it into HSBC and expect it to work. I split recruiting into six types. On the left you've got agency recruiting, on the right in-house talent acquisition. Then you've got volume, experienced hire, and leadership. On the agency side you can automate a little less, because they're typically asked to find people the employer couldn't find themselves. As you go from volume to experienced hire to executive, the opportunity to automate gets less and less. Amazon has famously automated processes where the candidate first meets their manager on day one. One big distribution company has a process where day one in the job is effectively the assessment. If it's a transactional requirement, say people who are going to turn up and drive in a peak period, you may not need to see them face to face, as long as they've gone through the checks, because motivation and values matter far less there than health, safety and a driving licence. You can cross-check those in an automated way.

Robert Newry: And I remember research asking candidates whether they want to meet a recruiter in the process. Overwhelmingly, no.

Adam Gordon: That's another thing we over-index on in TA, "they want to meet us." They actually don't. In experienced hire, do they want to meet a recruiter? No. Do they want to meet the person who'd be their manager? Probably yes. At leadership level they much more want to meet a recruiter, because they can have independent conversations they might not have with their would-be boss. That individual acts as a broker, handling a relationship on both sides, like a football agent, and the stakes are much bigger. On the in-house side there's more opportunity to automate, because you have budgets for candidate experience, talent nurture through your CRM, the career site, things the agency doesn't have the same access to.

Robert Newry: So we've got to think about recruitment in segments rather than as a single entity. Taking that perspective, everyone at the moment is coming up with an AI tool or feature. If you add the 1,200 from Randstad to the AI features ATS providers are pushing out, it feels like a gold rush. What's your perspective on the number of tools and features coming out around AI?

Adam Gordon: I've got quite a lot to say about this. Nearly everything is doing one of three things. The first is transcribing meeting notes and turning them into follow-ups and prompts. I think that's really useful. The second is sourcing, identifying people on the internet who might do the job. Nearly everything I've seen there looks exactly like the tools from 2016. It was advanced Boolean searching with keyword matching and an algorithm, surfacing individuals with a match score based on how close they are to the job description. That technology has been around for some time. The difference is a lot of what's coming out today has "AI" written all over it, but when you get into what it actually does, it's not doing anything dissimilar to 10 years ago. I'm sorry to be cynical, but some are raising huge amounts of money and I look at it and think, this is the emperor's new clothes. The winners will be the ones who tell the best stories and do the best marketing. The third area is assessment, which you know far better than I do. But I look at some of these AI assessment tools and half of them I immediately think are unethical.

Robert Newry: Unethical in what way, asking people questions they shouldn't be asked, or just not validated?

Adam Gordon: Partly that, but also things like assessing people according to the hesitancy in their voice. The amount of implied bias in these is really high. Finding people isn't the biggest problem; the biggest problem is in assessment. Someone with autism is going to come across very differently when interviewed by an avatar than someone who is neurotypical, and I don't think most of these things have those filters baked in yet. This is the bit I find fascinating: the difference between data science and social science. Data science just looks for correlations, this word is in the job description, this word is in their profile, they must be a good match. Social science asks what about people who refer to things differently. The story I give is personal. My daughter finished her PhD at Imperial, clearly very high cognitive ability, and wanted to be a project manager. She put her PhD on LinkedIn but didn't use the words that implied she had project management skills, because she didn't know what they were. Like many women in the research, she only put up words she felt she could defend, whereas men are more comfortable putting up words they aspire to. So she wasn't getting matches. I helped her adjust the words. She's now a project manager at the Crick Institute, on one of the largest cancer research programmes in the UK. She could fill a profile with the right words, but she'd never have got that opportunity through keyword matching. Where's the science behind some of this? The people winning are the ones telling the best stories, not necessarily the ones that have been tested and challenged.

Robert Newry: So how would you advise a TA leader to look under the hood, if they don't have the expertise?

Adam Gordon: Build a network of peers if you don't have one, and ask what they recommend, WhatsApp groups, LinkedIn connections, ask for recommendations. And, ironically, you could go onto ChatGPT and say, "I don't know what I don't know. What questions should I be asking to understand if this will help me achieve my tasks, and if it's ethically sound?" That's a great place to start. But I'm nervous about the amount of lying in marketing.

Robert Newry: Really? You think there's a lot?

Adam Gordon: Masses of it. It's not just mistruths or untruths. It's proper lying, and it's misleading. I can spot it a mile away. When product marketing doesn't match product capability, I can detect it quite quickly.

Robert Newry: Do you think that's people overselling something they think is coming later, selling a projection? Or is it that they haven't got anything exciting, so they tell a big story to break through the noise?

Adam Gordon: It all comes from the venture capital model. With each fund they invest in perhaps 20 companies, expecting 18 to fail and two to return the fund several times over. They want rapid growth and hype. The founders have done an amazing job to raise investment, and now they need it to work almost at all costs, because the stakes for their lives are extremely high. So the pressure runs from the VCs to the founders to the go-to-market leaders, and then to product marketing, demand generation, SDRs and account teams. Of the 1,200 companies that approached Randstad in the last year, maybe 200 will succeed and a thousand will fail. Around 30 will be very successful. It's worth understanding that maths when you think about what to invest in. My own principle, which is possibly anti-competitive, is that I market what I can achieve today. Marketing what's coming in three months is almost acceptable, especially with a caveat. Marketing what's at least six months away as if it's here is just lying. And on that note, I'm seeing a lot about agentic AI.

Robert Newry: That seems to be the buzzword of the moment. What does agentic AI mean? Is it fundamentally different, is it transformational, where are we with it?

Adam Gordon: I was in a room about 10 days ago with 30 UK-based resourcing leaders, and I asked how many could confidently describe what agentic AI means. About 40% put their hands up, which was higher than I expected, but at least 60% could not accurately describe it, even though it's the buzzword of the moment. In simple terms, an AI agent is a software program that can complete tasks autonomously after it's been told the instructions. The autonomous bit means it can make some decisions, use some initiative, but it's still achieving a task it was set up to achieve. A few things about it. There's a narrative that AI agents should be considered almost the same as employees. How I understand it, that's product marketing from AI agent companies, infusing the idea that "I can get rid of Sarah and John if I have this agent." They possibly can, but they're mistaken if they think the agent is equivalent to a human. It completes tasks. There's a narrative that they don't take holidays or sick days. Well, they do go wrong. They need babysitting, a high degree of monitoring. The agent is set up on preset criteria, so if one of those changes, because someone updated a feature or changed a bit of code, the chain falls apart.

Robert Newry: And that criteria could break because of a change in a large language model, a change in your systems of record, or another process changing.

Adam Gordon: Exactly, so they need constant supervision. If I was running a TA team of 20, I might get it to 15 with AI agents, but I would need an engineer in my team, and I would need a talent product manager. The product manager is a recruitment operations person who designs what the workflow needs to look like and continuously monitors and enhances it. The engineer continuously looks after the technology. There's hardly anyone in talent acquisition who could do the engineering job today.

Robert Newry: So there's a lot of misconception that an agent can operate on its own, and that it will be easy for one to replace a human. Where do you think the function of TA needs to structure itself now?

Adam Gordon: Split it by agency and in-house, by the three levels, and also by supply and demand, some industries have an oversupply of qualified people, others have in-demand talent that's hard to reach. On the shape of the team, it's been changing a lot over five years. We need fewer talent sourcers, fewer interviewers, fewer coordinators scheduling interviews, and most teams have already rationalised those. To become consistently more productive, TA teams have been emphasising recruitment operations, the policies, processes and technologies, designing the experiences and workflows and doing the change management. We've not really valued or budgeted that type of role. It's been "TA needs some technology, let's go to IT", or the ATS provider sells to IT and says flick on this feature. It's not been about the process, the workflow, the configuration. TA needs that skill in-house rather than borrowing it from IT. Every white-collar function does, finance operations, marketing operations. We're a little behind because we're smaller, but these teams are growing. If your core skill is interviewing or talent sourcing, get much more in tune with recruitment operations, because that's where a lot of the jobs will be. The other area is candidate experience and employer branding. Despite shrinking headcount, TA is equally, if not more, important, because with less fixed headcount each person needs to accomplish more. Employer branding, conveying the message you want the market to think about your organisation, and designing the experience from the first moment someone meets your brand, will only get more important.

Robert Newry: One guest earlier this series talked about making the recruitment process frictionless for the recruiter, but not effortless for the candidate. A lot of organisations, Amazon among them, went for one-click apply, and that's coming back to bite us. People are starting to think about putting a bit more friction back in to see if someone is genuinely motivated. Are you seeing that?

Adam Gordon: A hundred percent. The focus on recruiter experience is extremely valuable. If I take an organisation that is the employer of choice in a given area, they can make people do a decathlon as part of the assessment, and whoever completes it gets considered. In areas where they're competing with other employers of choice, they'll have to roll out the red carpet as much as they expect effort from the individual. It might be more 50-50. It's exactly like dating: if you're a 10 out of 10 in every aspect, you can be choosy. Very few are. Sometimes you put in more effort, sometimes less, and it's the same with recruiting.

Robert Newry: One thing that keeps coming up is, how do we keep the human in the loop? It seems to me that when people ask that, what they really mean is, how do I protect my job? What's your take?

Adam Gordon: It's a defensive term from people nervous about their jobs being overtaken by technology. It's a natural human positioning. There are just some jobs where the human doesn't need to be in the loop and is actually getting in the way. Where does the human need to be in the loop? For most jobs, when it comes to determining who's going to be interviewed. It would be desirable for a human to review a CV or an application, or at least set the criteria, because nobody wants to go and look at a thousand CVs. So what are the knockout questions and knockout phases? If we've got one job and 3,000 applicants, the human doesn't have time to review more than maybe 50 CVs. So we're going to have to reject 2,950 automatically.

Robert Newry: And there's a misconception that candidates aren't being screened out by AI.

Adam Gordon: They absolutely are, and I don't know why we pretend otherwise, whether it's AI or a simpler check, do you have the right to work, what's your degree. That's always been the case. What TA should do more of collectively is explain the challenges, why this is happening. They don't have enough people to manually review 3,000 applicants. And be honest that if you get a human to look through 3,000 CVs, very quickly they start making shortcuts, we can't process everything. You end up with ridiculous shortcuts, do they play for a sports team, oh I like chess, or "I'll just cut 50% because they're unlucky." So it's about finding the fairest possible way to get from an enormous number down to some you can really focus on.

Robert Newry: So what's your advice to candidates?

Adam Gordon: You've got to work out how to hack the system. I don't mean anything fraudulent. Get yourself known to the recruiter and the hiring manager by connecting with them. Write blogs about your subject, get yourself on stage, on podcasts, build your personal brand. If it's your dream job and you're one of 3,000, you've got to find a way of standing out. I once hired someone who wouldn't have made my interview shortlist based on the content of her CV, but the way she'd designed it was beautiful, I'd never seen anyone do that, and it showed real initiative and creative thinking. And Pete, who works in my team at Poetry, wasn't someone I was hiring for. He sent me a video of a software overlay he'd built for Poetry himself and recorded how it would work. I sent it to my co-founder Mike and said we need to talk to this guy. We hired him at least three months earlier than planned, because we were so impressed with how he'd hacked his way in front of us. The spray-and-pray approach isn't the answer. Focus on where you want to go, and then do everything you can to demonstrate why you're better than everyone else.

Robert Newry: Adam, it's been fascinating, as I knew it would be. What you've shared about how TA leaders need to think about using technology, how they implement it, and how they organise themselves to stay ahead has been incredibly useful. Thanks very much for coming on.

Adam Gordon: Thanks so much for having me. It's been a real honour.

Robert Newry: Let me pull a few threads out before we close the series. One thing I thought was fantastic was Adam's thinking about the role of the TA function going forward, not just implementing technology, but asking what the purpose of TA is and how it configures and adapts its workflows for a different world. I also liked his point on "human in the loop", and thinking about where and what that really means, because where TA adds value is going to be different going forward. Those who succeed will be the ones ahead of the curve, not the laggards. And the last thread was candidate experience: we have to put ourselves in the shoes of candidates, understand the pressure they're under and how they need to stand out, and think especially about those on the edges of the candidate pool who don't have that support, so that people who've been marginalised aren't left further behind. If anything today resonated, reach out to me or Adam on LinkedIn. And if you've found this series useful, please like and subscribe. We'll be back with the next series soon. Thank you.