Are we overestimating AI Agent advancement?

While AI agents are transforming industries, the gap between potential and practicality remains, hindering widespread adoption and integration, writes Jeffrey Tobias

When we think of AI agents, we tend to think of their “promise”.

AI agents offer an undeniable and inevitable transformation of workplaces, industries, and, in some parts, the world. Alexa and Siri are two of the most famous AI agents in popular culture today, and they have transformed how we interact with our phones, in our homes, and in society.

However, despite this incredibly successful example, I’d argue that we may be overestimating its advancement so far.

Now, I’m not an AI-sceptic. Far from it. I have a PhD in Computer Science, I use AI in my work every day, all day, and I stay current through articles and podcasts. So when I see bold predictions for AI agents in the next 12 months, I have to ask – are we there yet?

Take Google’s Project Astra, still without a public beta or release date. AutoGPT, once hyped as the future of autonomous agents, has largely fizzled due to reliability issues and a tendency to loop or hallucinate. Amazon’s Rufus, their AI shopping assistant, launched to criticism for offering little beyond basic chatbot capabilities. And xAI’s Grok, billed as a reasoning breakthrough, has repeatedly delivered inaccurate or misleading answers.

If you look at Gartner’s Hype Cycle, perhaps we’re moving from the ‘peak of inflated expectations’ to the ‘slope of enlightenment’.

In short, as much as I’d like to believe we’re already there, I know we’re not.

Vibing with agents is not nearly as easy as people say it is

There’s been a lot of hype recently about ‘vibecoding’ and vibing with AI. It shows how people are talking to AI agents and creating work in minutes that previously would have taken weeks to create.

Well, I’ve tried something similar.

At the Strategy Group, we thought we'd delve into building an example of agents. There is a lot of hype around agents and the work they can do: similar to vibing, they’re meant to be revolutionary in terms of work impact. A task that used to take hours should now take minutes. And, certainly, we're heading down that path, but it's not that simple at the moment.

Read more: The strategic impact of AI on business transformation

The reality is that a task to create an operating set of agents that I believed would take me 30 minutes ended up taking me days to get up and running. I quickly learned that this would not be simple. We’re not really at the stage where it’s a simple command to spin up an agent, define the task, and it’s done in minutes. At this stage, you still need to have some understanding of coding, spot basic errors, and redirect where possible.

That’s not a skill everyone possesses, and this is where I think we fall short currently with AI agents in general. We are promising accessibility to everyone, but this level of tool is not ready for general use. It’s there and ready for those with the skills necessary to use it effectively.

It’s here that we need to bridge the gap.

There’s a gap between potential and practicality

I’d argue that the narrative of AI agents revolutionising industries is compelling, but it often ignores the practical barriers to implementation.

One of the most persistent misconceptions is that these agents (and AI) will completely transform industries within a short timeframe. In reality, historical precedent suggests otherwise. Consider the adoption of personal computers or the internet, which first came around in 1971 when the first email was sent, but it didn’t gain any notoriety really until the invention of the World Wide Web in 1989. That’s nearly 20 years between one big event and another, and there were countless developments in between.

And then again, these innovations took decades to become fully embedded in business operations. We might use the internet without thinking now, but think how long it took to get here?

In a similar vein, Microsoft first announced it wanted to democratise AI in 2016, nine years ago. With the mainstream use of Chat GPT coming into play in 2022, you’d be forgiven for thinking that this is an overnight success.

It’s not. And it won’t be.

AI, and in particular AI agents, will likely follow a trajectory similar to that of the Internet, requiring incremental improvements, workforce adaptation, and regulatory clarity before reaching mainstream integration.

In some ways, I believe that this is how it should be, but I also want to encourage organisations to adapt sooner rather than later.

The myth of immediate AI agent disruption

There is a mythical tendency about how imminent AI disruption is, particularly with the rise of AI agents.

However, from working with our clients building their AI strategy, I’ve found that many companies face roadblocks when attempting to integrate AI, let alone AI agents, into their processes.

For example, in my article AI is Democratising. What’s Next?, I explained that fear stops companies from innovating. The 2024 Cisco Privacy Benchmark Study, for example, found that 27% of organisations have banned generative AI tools due to security and privacy concerns. A further 61% have limits on how GenAI tools can be used by employees.

Subscribe to BusinessThink for the latest research, analysis and insights from UNSW Business School

So, I’ll continue to question: If companies react to AI with fear rather than strategy, will global innovation continue to stall?

From a business standpoint, AI agent adoption is not simply a matter of acquiring the technology but requires a transformation in three crucial elements: strategy, operations, and culture. It needs to be strategically implemented without a focus on immediate ROI but instead embedded into daily operations. Entire teams need to adopt it daily, with a culture of embracing new innovative tools designed to enhance workloads.

And all of this takes time.

We need less hype, more realism

AI agents are powerful, but they are not a silver bullet.

The conversation needs to shift from technology vendors' speculative futurism to grounded realism. While AI’s capabilities expand, the opportunity for organisation-wide agent deployment is truly exciting; however, the barriers to widespread agent adoption remain formidable. The true challenge lies not in making AI smarter but in bridging the gap between technological potential and real-world implementation. “Show me the use case" is what I hear all the time.

Rather than overestimating the agent’s trajectory, we should focus on what it truly takes to make this work in practice, such as change management, leadership in a whole new world, and how we will retrain the literally millions of workers whose roles will need to be redefined.