Everyone Is Talking About AI Agents. I Think They're Looking at the Wrong Thing.
Over the last few months, I've been hearing the term "AI Agent" everywhere.
Every demo is an agent. Every startup is building agents. Every investor deck seems to have at least one slide about agents.
Maybe it's just the natural progression of the AI hype cycle. We get excited about something, oversell it for a while, and then eventually figure out where the real value is.
But I do think something genuinely important is happening beneath all the noise.
What's interesting is that most of the conversations I'm having with businesses aren't actually about agents.
At least not directly.
A typical conversation goes something like this:
"Should we use OpenAI or Anthropic?"
Or: "Do you think GPT-5 is significantly better than GPT-4?"
Five minutes later we're discussing something completely different.
Usually it's a process problem.
Information scattered across different systems. Support teams manually updating records. Operations teams copying data from one platform into another. Employees searching through documents because nobody remembers where anything is stored.
The funny thing is that the AI model is rarely the bottleneck.
The workflow is.
That's why I think many people are focusing on the wrong aspect of AI agents.
The interesting part isn't that they're intelligent.
The interesting part is that they can participate in a process.
That's a very different conversation.
Take something simple like customer onboarding.
- A new customer submits a form.
- Someone checks the information.
- Someone creates an account.
- Someone updates a CRM.
- Someone sends a welcome email.
- Someone schedules a follow-up.
None of those tasks are particularly difficult. Yet they happen thousands of times across businesses every day.
Most discussions around AI focus on making one step faster.
An agent becomes interesting when it can handle multiple steps across the entire workflow.
Not perfectly. Not autonomously forever. But well enough that humans are only involved when they're actually needed.
That's a much bigger shift than generating a slightly better email.
I also think some of the expectations around agents are getting a little ahead of reality.
A lot of the online discussions make it sound like businesses will soon replace entire teams with autonomous software.
Maybe one day. I'm not betting on that happening anytime soon.
What I am seeing is companies looking for ways to remove repetitive work.
Not jobs. Work. There's a difference.
Nobody enjoys spending an hour updating records in three different systems because those systems don't talk to each other.
Nobody enjoys searching through twenty PDFs to find a piece of information that should have been accessible in thirty seconds.
Nobody enjoys copying information from one dashboard to another.
Those are the kinds of problems businesses actually want solved. And honestly, they're not new problems. We've had them for years. AI is simply giving us a new way to approach them.
Something else I've noticed recently is that many businesses are no longer asking whether AI works.
That conversation has largely ended.
A year ago people were experimenting. Today they're budgeting. They're planning. They're asking about integration, security, governance and ROI.
That's usually a sign that a technology is moving beyond curiosity and becoming infrastructure.
Not infrastructure in the data centre sense. Infrastructure in the business sense. The kind of thing that quietly becomes part of how work gets done.
Maybe that's where AI agents ultimately end up.
Not as digital employees. Not as magical autonomous workers. Just as another layer of software that helps organisations operate more efficiently.
That may sound less exciting than some of the headlines.
But if it happens at scale, it could be far more impactful.
And that's probably why I'm paying attention.