AI Agents: Hype, Hope, and the Path to Practicality
Are AI Agents the Next Big Thing or Just a Shiny Distraction?
Welcome to FullStack HR, and an extra welcome to the 48 people who have signed up since last time. If you haven’t yet subscribed, join the 8400+ smart, curious HR folks by subscribing here:
Listen to this episode on Spotify.
Happy Tuesday,
This is the second article this year and I’m still exploring some of the trends I talked about before Christmas. And this was a trend that I didn’t include and that many in the webinar pointed out that they would like to hear about. Hence here we are, talking about AI Agents.
I’ve written about agents in the past so I do have a view on it. Or I had a view and I’ve probably reverted that view a couple of times.
Heck, even while writing this article I’ve reverted and updated my view on agents and that is part of the problem.
But my main thinking here is while AI Agents are insanely cool, they have a bit of way to go before being ready to be fully launched, operational and valuable inside organisations.
Speaking of agents, as part of me learning more in-depth about agents, I wanted to really work with an agent on an idea I’ve had for a while. The idea is centered around a tool that helps organisation to adopt to the EU AI Act, with practical insights.
I’ll launch that tool very soon and it will of course be free.
Interested? Sign up here and I’ll let you know as soon as it’s available!
AI agents are the new darling of the (tech) world. Searches for "AI agents" have skyrocketed over the past year, and the internet is awash with headlines proclaiming them as the future of work. Every other blog post, panel, and newsletter seems to bring up the same question: are AI agents the next big thing?
And sure, I get the hype. I’ve even used some over the holidays, like Replit’s agent, and I’ll admit, it was impressive. But still, I can’t help feeling a bit... conflicted.
On one hand, I absolutely see the potential. A good AI agent could transform how we work, automate tedious tasks, and free us up to focus on more meaningful things. But on the other hand, the concept feels a bit nebulous. What even is an AI agent? Are tools like ChatGPT or Claude agents? Or are they just pretending to be?
And here’s where I struggle: while we’re seeing all this noise about AI agents, most organizations haven’t even figured out the basics of generative AI yet. So is this the next big thing—or just a shiny distraction?
What even is an AI Agent?
With all the buzz, you'd think we'd have a clear definition of what an AI agent actually is. But that's part of the problem—the concept isn't as straightforward as it sounds.
Chip Huyen, a leading voice in the AI community, offers a useful starting point:
"An agent is anything that can perceive its environment and act upon that environment."
This makes sense in theory, but in practice, agents are far from simple. Their capabilities depend on two things: the environment they operate in and the tools they can use. For example:
A coding agent like Software Engineering-agent works in a computer terminal, handling tasks like navigating repositories, searching files, and editing code.
General-purpose tools like ChatGPT or Claude can act as agents by helping with SQL queries, managing customer accounts, or even planning a trip.
But as straightforward as the definition might seem, the reality of implementing agents is full of challenges. As Huyen points out, agents need powerful tools and well-defined environments to deliver on their promises—and even then, they're prone to inefficiencies and planning errors.
The hype is real—but so are the challenges
So, we've established that AI agents are the current hyper-trend in the tech world.
But beyond the buzz, it's clear that agents are still in their infancy for most organizations. Making them work isn't just a matter of deploying an AI model—it requires foundational work, such as:
Documenting processes: Agents need clear workflows to automate tasks effectively.
Building toolkits: Without access to the right tools, agents are severely limited.
Preparing teams: Many organizations are still grappling with the basics of generative AI, let alone deploying agents.
And then there's the practical issue of interaction.
Most agents today rely on text-based instructions, which can be clunky and slow.
And let’s trye a quick experiment: close your eyes for a moment.
Now open them and take in everything around you.
In just seconds, your brain processes an incredible amount of visual information—context, relationships, and next steps.
You don’t need instructions or someone spelling it out for you; it’s intuitive.
Now compare that to how we interact with most AI agents today. Instead of seeing or intuitively understanding a situation, you have to describe it in detail using text. Every instruction has to be spelled out, often in painstaking clarity, for the agent to act.
It’s not just slower—it creates a bottleneck that limits how effectively we can work with AI.
Until agents can “see” and process information as naturally as humans, this reliance on text will remain a major barrier to their usability and adoption.
Contrarian take: It’s not all hype—some are already doing it
While it's easy to be skeptical, some companies are proving that AI agents aren't just buzzwords. Nvidia is a prime example.
Their CEO, Jensen Huang, recently shared a bold vision:
"We have AI agents helping design chips—Hopper wouldn’t be possible, Blackwell wouldn’t be possible, and don’t even think about Rubin. [...] I’m hoping that Nvidia someday will be a 50,000 employee company with 100 million AI assistants [...] some of them digital and some biological."
If this holds true, which I think it does, Nvidia isn't experimenting with agents; they're using them to design chips, verify processes, and develop software.
For them, agents are an integral part of how they operate.
And it's not just Nvidia. Chip Huyen notes that the economic potential of agents is enormous, capable of automating tasks like market research, data entry, and even interviewing candidates. Agents like SWE-agent are already performing complex coding tasks by navigating repositories and editing code.
But let's be honest, I think companies like Nvidia are outliers. They have the resources, technical expertise, and infrastructure to make this work.
For the rest of us, it's a reminder that agents won't succeed without first laying a strong foundation.
So, what can HR do right now?
If you're in HR (which I assume most of you readin is), you might be wondering:
"What does this mean for me?"
While agents might not be ready for everyone, there's still plenty you can do to prepare:
Understand generative AI: Before you even think about agents, get comfortable with the basics. What can tools like ChatGPT or Claude do for your team?
Map your processes: Create detailed workflows to identify tasks that are repetitive or rules-based. These are the best candidates for automation.
Experiment small: Start with simple use cases and learn from them. For example, collaborate with platforms like klicka.ai to explore early-stage agent solutions.
By focusing on these steps, you'll be better positioned to leverage agents when the time is right.
So what do I really think?
I think AI agents are fascinating and to some extent amazing. As mentioned, working with Replit, which I deem to truly be an agent, is mind-blowing. You have an idea about something an Replit executes. It’s insane and I do think that we’ll see even more of this in the future.
So I do believe they have the potential to transform how we work. But let's not get ahead of ourselves. For most of us, agents are still in the early stages—full of promise but requiring a lot of groundwork.
That said, companies like Nvidia show what's possible when you get it right. And while most of us aren't there yet, it's a vision worth keeping in mind as we build toward the future. And I do think that there will be A LOT of talk about agents in 2025 but I think there will be limited economical impact from agents in 2025.
Sure, a lot of companies will experiment with agents but for the majority, I think we still need to catch up on and nurture the GenAI-tools we have available.
So, what's your next step?
Start small. Focus on understanding AI, mapping processes, and experimenting with the tools you have. Because when agents are ready, you'll be ready too.
I think there is a great use case for HRBPs who haven’t yet got there head around data analytics. Or at least understanding how data is organised in systems to answer questions. Ai agents may help HRBPs test assumptions about the organisation such as “increasing engagement scores reduces attrition”. Ai agents can quickly test these hypothesis without taking up huge time of limited HR Analytics resources.