If you haven’t already, subscribe and join our community in receiving weekly AI insights, updates and interviews with industry experts straight to your feed.

DeepDive 

Your weekly immersion in AI  

Let’s start with a question: would you hire an assistant without ever writing a job description? 

Even if they were the most capable assistant in the world, they wouldn’t know what you wanted from them. 

And if you think about it, this is kind of where we are with AI agents. Over the last year, the conversation has focused on what these systems can do. Build software, manage workflows, coordinate other AI systems; complete more and more complex tasks with minimal human intervention.

But maybe we’ve been missing the critical question – and really, we should be asking what we actually want AI agents to be.

Riding the wave of expectations

According to Gartner's 2026 Hype Cycle for Agentic AI, agentic AI has reached the ‘Peak of Inflated Expectations’. To be clear, that's not a verdict that the technology is overhyped or destined to disappoint. Instead, it reflects the extraordinary level of excitement surrounding the field – and the reality that the term ‘AI agent’ now encompasses a wide range of technologies at very different stages of maturity.

Gartner's advice is pragmatic. Rather than assuming every product labelled as an AI agent is equally capable or ready for deployment, organisations should evaluate these systems in terms of governance, security, cost, and organisational readiness, recognising that different agent technologies will mature on different timelines.

Because hype can still influence expectations just as much as technology itself. When every new product promises an autonomous digital workforce, it's easy to start measuring progress by one metric alone: how much more the agent can do without us.

But is greater autonomy always the goal?

The question beyond the tech

That was the question that caught our attention in a recent interview with MIT Professor Phillip Isola, published on MIT News. 

He offers a wonderfully simple definition: "Agentic AI is AI that takes actions in the world."

So generative AI creates, and agentic AI acts. 

Once an AI system begins taking actions without being micromanaged by us, we need to stop focusing only on capability and start looking at delegation. What do we genuinely want and need to delegate? And if AI is making decisions for us, what knock-on effects will that have over time? 

In the interview, Isola argues that the real challenge isn't building agents that can do more. It's deciding which decisions we want to hand over, and which should remain firmly in human hands.

Some answers are obvious. Not many of us would object to an AI agent automatically sorting receipts or scheduling meetings. But as agents become more capable, the boundaries become less clear. 

• Should an agent decide which job candidates make it to interview?
• Should it negotiate on your behalf?
• Should it decide which medical information is most relevant before a clinician reviews it?

These are human questions, not engineering ones. 

Designing collaboration, not replacement

One of the most interesting ideas emerging from both Gartner and MIT is that the future of agentic AI isn't really about autonomy – it's about collaboration.

The most successful organisations are unlikely to be those that remove humans from every decision. They'll be the ones that carefully design where human judgement adds value and where intelligent automation genuinely reduces friction.

That requires a shift in thinking. Rather than asking if an agent can perform a task, we need to ask whether it should. 

What comes next?

AI agents promise a future where software works alongside us. That's an exciting prospect. But every technological leap comes with choices about how we want to work, create and make decisions.

What do you want AI agents to be? 

Open this newsletter on LinkedIn and share your perspective.