AI Agents Can Do Anything

Arham: Today, we’re going to be talking about Yann’s latest article on Substack, which you can read. It’s about how AI agents are not just chatbots. Hi Yann, can you talk a bit more about that analogy in your article: chatbots being like a microwave and AI agents being like a chef? Could you open this up a bit?

Yann: A chatbot is first confined, you know, to a window. You can train it with your own information, and it can do a few things, for instance, with Dialogflow and so forth. But in the end, the chatbot is limited, right? Whereas an AI agent can do anything.

We’ve seen it; it has its own platform. On the line graph side, it has access to services, to APIs, to GCP servers, and it can do workflow automation. So, if you want, the agent has a lot more superpower to do things. I think the chatbot was the beginning, and now we have the AI agent, and there’s so much more that we can do with that.

Arham: I want you to talk a bit more about some successful implementations. You also talked about AI agents: Credit Agricole, Mayo Clinic, even Zazmic itself had some pretty interesting AI agent implementations. Could you talk about the kind of disruption you see it causing industry-wide? 

Yann: We’ve done end-to-end projects that completely take care of serving a customer, enabling a customer, and making a customer ready in production. This is something you cannot do with a chatbot. With a team of agents, it enables a client to purchase some hardware via a portal, to configure that hardware in the portal via an agent, and then to set it up in production in their premise, going up and working. This would have never been possible two or three years ago, and it all happens using agents without hiccups.

So, very complex tasks—from software to purchasing an item, to configuring a piece of hardware, and launching it in production—which is a complex thing. Yeah, there’s so much more. That’s the chef in the kitchen right there for you.

We’ve done projects that completely, end-to-end take care of serving a customer, enabling them, and getting them ready in production. You just can’t do that with a chatbot. Our team of agents enables a client to purchase hardware via a portal, configure that hardware in the portal through an agent, and then set it up in production at their premise, getting it up and working. This would have been impossible two or three years ago, and it all happens using agents without a single hiccup.

So, it’s about handling very complex tasks—from software processes to purchasing an item, configuring a piece of hardware, and launching it in production, which is a really complex thing. Yeah, there’s so much more. That’s the chef in the kitchen right there for you.

Arham: You mentioned a three-part process for running AI agents: planning, learning, and then acting. I think even with some of the newer models coming out, we’re seeing how they’re able to reason with their own responses, which is fascinating.

So, applying this to businesses—because I’m sure there’s no business that doesn’t want to shift into using AI agents at this point—from a change management perspective, how do you think a business can best manage this shift?

Yann: Hard. I think the change management part is hard. It needs to come from the top. You’re enabling your team to go 10x in productivity and sometimes replace some of the tasks your team does. It’s not about replacing people, but it’s about enabling them to do much more in their day-to-day work. So I think that’s one part.

And to address your second part, it’s true that the agent now can think, can reason. Within the scope of what he or she has to do, the agent is basically able to be pretty independent. So I think that brings a lot of sophistication into this agent or team of agents solution that you can put in place in your company.

Arham: As you said right at the end, stop missing out and start doing when it comes to getting AI agents in the organization.

Yann: The time is now. I think everybody’s doing it. People are missing out on productivity if they don’t. It needs to come from the top, otherwise people will always find a reason not to use them.