The secret to successful AI in hiring? Stop overthinking it

Business Insider

01-05-2025

Good news for any talent leaders feeling overwhelmed by AI's developmental onslaught: "If they're reading this article, they shouldn't worry," Trey Causey, Indeed head of Responsible AI, said. "Anyone even considering using AI is already ahead of the curve."
In this interview, Causey cuts through the hype and breaks down what you need to know about the latest AI technology in hiring — generative AI (GenAI) and agentic AI — and the impending arrival of artificial general intelligence (AGI), or superintelligence. The below conversation has been edited for length and clarity.
GenAI, such as ChatGPT, has been around for a while. How does agentic AI differ, and what use cases do you see in hiring and retention?
Causey: What makes GenAI unique is that, as the name suggests, it generates new outputs and content based on a prompt or a set of instructions the user provides. That could be a job description, handling customer service inquiries — anything where you need ideas.
Agentic AI is the natural next step. Many current GenAI systems are chat-based, and everything happens within the confines of that chat. But AI "agents" are like assistants that can do things for us. It's their independent actions that separate agentic AI.
For instance, you can set up an agent to review the daily applications to an open role, summarize those applications, identify the candidates who have the skills and experience you're seeking, and generate a report that orders those candidates with an overall summary to review at the end of your day. Then you can tell it to craft and send a personalized message to each candidate you approve, invite them to a screening call, and alert you when they've responded.
Before, the AI wouldn't be able to interact with other systems; it wouldn't be able to go get those resumes unless you somehow had put them all into the context that it had access to. But with agentic AI, you can keep adding steps to this chain and it'll work in the background while you take care of other things. We're working on streamlining all of that on Indeed.
Are there misconceptions or pitfalls unique to agentic AI that users should be aware of?
Causey: AI systems are prone to flaws and mistakes. Just because it's the next evolution doesn't mean it's perfect.
These agents are designed to take action independently, but that means the cost of mistakes is higher — if you reach out to a candidate, you can't take that back. It's important to be intentional about what you enable AI agents to do, making sure you have a way to review the tasks and outputs. It would be misguided to immediately delegate all of your work to an agent right now.
It's like the early days of self-driving cars: You still need your hands on the wheel.
Indeed's recent global report reveals both employers and job seekers support skills-first hiring, but limited time and resources are barriers. How can agentic AI help?
Causey: In the transition to skills-first hiring, the biggest puzzles are:
How do we know that job seekers have the skills we need?
Do the job seekers even know?
How do we verify both sides of that equation in a way that both the job seeker and the employer trust?
Imagine having an AI agent automatically look at resumes and not only extract the skills listed, but also use information on the back end to map other skills to the positions candidates held previously. It could even follow up with the job seeker to say, "These skills aren't on your resume. But from your experience in jobs A, B, and C, you might have them. Would you like to take an assessment?"
Automating that back-and-forth avoids ruling someone out for not using the "right" language on their resume and prevents the recruiter from submitting someone for review only when they have the time to follow up on those skills and wait for a response. The assessments close the trust gap so the employer can quickly verify the essentials and get to interviewing.
Indeed's global survey also shows that workers increasingly value learning and development opportunities when choosing employers, even over pay. How can employers use AI in L&D to better attract and retain talent?
Causey: AI opens up a lot of opportunities to make L&D on-demand to employees at scale and at a relatively low cost. It can construct personalized learning plans and study materials, then create an assessment to see how well you're learning and provide opportunities to practice at your own pace.
But there are still social elements. It's difficult to stay accountable with online learning. Maybe it's nine at night, you just put your kid to bed, and you really don't want to learn Python right now. That's where a manager can support and motivate. The human component is always key to success.
Can AI also help with work wellbeing?
Causey: While we don't want to create a surveillance culture, I do think it can be useful if a manager gets overburdened and may not notice one of their team members is becoming disengaged.
For example, imagine you've collected data on absences. An agent can regularly compile a report to identify employees who might need a break. There are so many ways we can aggregate data to make it easily accessible and actionable.
How does artificial general intelligence, AI's supposed next evolution, differ from the other forms of AI we've discussed?
Causey: Artificial general intelligence is basically a system or set of systems that can outperform humans at any task. But there's no agreed-upon definition of what that looks like, so some jokingly say it's "whatever we don't have yet." It's more of an academic debate at the moment.
Most of the large AI labs have been shortening their timelines for when we'll see AGI, including the engineers actually working on these systems. This has led to some proposed AGI nightmare scenarios that I don't find super compelling. Just because something is very intelligent or has the appearance of intelligence doesn't mean it can do everything humans do.
So what do employers need to know about AGI right now?
Causey: My hot take is they don't need to care. Regarding the macroeconomic implications of AGI, so many outcomes are equally probable right now that you can't do anything until there's more information. Whether or not AGI happens and when is much less important than what we're doing with the systems we have now.
Rather than spending time figuring out the right type of AI to use or where to use it, just start using AI in everything (within your company's policy and the parameters provided to you, of course). An experiment-driven approach lowers the stakes and relieves the pressure of perfectionism. Using AI is like anything else: If you don't practice, you don't get good at it.