AI Agents 101: The Shift from AI Copilots to Autopilots
The rise of AI agents is arguably the defining theme of tech in 2025, but the whole idea of "agentic AI" is overwhelming for many people. It turns out that agents are not really this big, super sophisticated, sci-fi kind of thing. They can actually be pretty pedestrian albeit incredibly powerful.
I spoke with Dave Hecker, who has a long resume as a CTO in Silicon Valley, for some clarification about the future of agents. He works with a wide variety of clients developing a range of agentic AI solutions. Our Q&A follows below, but first a bit of background.
Microsoft CEO Satya Nadella explained AI agents in the most basic terms at the annual Microsoft Build conference last month. "You can ask questions and AI assistants give us answers," he said. "You can assign tasks to agents and have them execute them."
Ultimately, we're talking about advanced automated workflows. They're designed to save time and resources completing tasks of all levels of complexity with varying degrees of autonomy, which have traditionally been managed by humans.
The evolution from AI assistants to agents is synonymous with the shift from large language models to large action models, or from AI copilots to autopilots. The other comparison between AI assistants and agents is that assistants are never going to replace any human jobs. It's the autonomous agents that are reducing and/or redistributing human labor in various sectors of the workforce.
Many people experienced with AI will debate the difference between automated workflows and true AI agents but that delineation seems to be evaporating. It's really more of a graduated spectrum with different levels of autonomy. Today, people are increasingly referring to any kind of automated systems with AI integrations as agents.
Either way, the business outcomes are proven. Agents are well established at the enterprise level. Salesforce launched its AI-powered Agentforce tool last year. "The Digital Labor Platform" is designed to help sales reps automate their prospecting processes.
Meanwhile, small to medium size organizations are embracing a growing stable of tools to develop automated workflows/agents, which are relatively affordable and don't require much coding ability. Make, Zapier and n8n are some of the more popular choices. Relevance and Relay have a growing fanbase, and ActiveCampaign is impressive for email automation with AI support. Don't expect plug and play functionality though, because they all require a certain level of expertise to work effectively and consistently.
A sample workflow/agent for a tourism organization or hotel could be developed to direct a new visitor/guest review or event planner inquiry into a CRM, then a Google Drive/MS OneDrive record and Slack channel, then an LLM like ChatGPT for summarizing and prioritizing, then another Google Drive/MS OneDrive record for the summary, which is then reproduced as an email to be sent to any number of internal executives — without any humans in the loop.
There are limitless scenarios for building agentic workflows.
Q&A: AI Agents & Automation
Greg Oates: Dave, how do you define what an AI agent is? And does it seem like more people these days are using "agent" and "automation" interchangeably?
Dave Hecker: We're going to stop using the word "agent" altogether pretty soon. I think it's going to be like the "information superhighway" and "world wide web" and stuff like that. I don't know what an agent is. The broad definition is that it's not just an AI that you can talk to or get information from or interact with, but it actually does a distinct task for you. Hence the term "agentic." But I think we're quickly reaching the point where an AI that doesn't do anything is kind of old hat, right? Everybody's enjoying talking to ChatGPT now, but it's often really just for fun. It's a bit of a parlor trick today for many people. But if I have ChatGPT reading my email and summarizing it, then I call it an agent.
And yeah, I agree with you. This distinction between agent and automation is not very meaningful, and it's very high level. I just think in the future, and even currently, AI tools are going to be able to do more and more things every day. And the idea of an agent versus non-agent is going to become sort of a silly semantic thing that we were involved in during the early days of AI. Soon, an AI will be judged really only by what it can actually do, and an AI that you just talk to I think is going to be pretty boring.
GO: So let's say a tourism organization or hotel wants to develop agentic systems. With tools like n8n, speaking from experience, there's a significant learning curve involved but the benefits are real. What do travel leaders need to know about these tools?
DH: A lot of these tech platforms do 90 or 95% of what you need. Then you're kinda stuck, right? They do almost everything, but then you need some custom code to make it do exactly what you want. Now, it's very possible to have that code written by ChatGPT, which is what I do most of the time. I have ChatGPT write the code, but being an engineer, I can understand the code and kind of make sure it's doing what I want it to do. So anyone looking at this should have some basic technical resources. You also want to be really conversant in the space and pretty clear on what these tools can and cannot do, and what's feasible and realistic.
GO: Is it true that one of the biggest challenges facing agentic AI is stacking all these different workflows and various APIs together? Which is why Anthropic's Model Context Protocol is all the rage now to help AIs and apps talk to each other nicely?
DH: Yes, what's keeping us from having truly agentic AI is the fact that things aren't very well integrated. You know, when you play with n8n, you've got this nice workflow engine that looks like a flow chart. Pretty simple. It's not so different than Visio 25 years ago. But it also has the ability to connect to the APIs of outside tools and that is the secret sauce. It's not always that easy to connect to an API though, and it's also often not that hard. So right now, any of these tools are filling an important gap, and they're allowing non-coders to do simple things and engineers to do pretty complex things much more quickly.
In two years, it's going to look very different. The API piece is going to be much easier to work with, so that won't be a big deal. But there will be thousands of more services and things that an agent can do and interesting use cases that aren't ready yet. Like right now, we see a lot of tricky AIs that can read a news story, make a video about it, and post it on Twitter. Which is very cool, but we all know that the videos aren't very interesting. It's just not commercially useful yet, but that's slowly changing and it's going to expand quickly. So, yeah, everything, everything's going to be an agent eventually.
GO: Will some of the platforms we're using to develop agents today become obsolete anytime soon?
DH: The answer is 100% yes. Some will be obsolete in two years, but it's kind of an interesting thing because, you know, I've been in technology for a long time. It's always been the case that everything we work on is soon going to be obsolete. That's nothing new. However, the pace has never been anything like this. About 15% of the code coming out of Silicon Valley was written by AI up until 2025. This year it's probably double that, and that's a massive acceleration. So yes, everything is going to change. But I think the dynamic that's always been in place will always still be in place. People who are competitive and effectively using the tools today are the ones who are going to keep up and know how to use the new tools tomorrow, and continue to do so as force multipliers.
GO: This is a bit general, I know, but how will agents impact the workforce in the next few years?
DH: Let's say one developer armed with AI can now do the work of three developers, which is debatably becoming the case. It doesn't mean that all is lost and everybody's out of work. It means that everyone is now competing at a higher level. You will need to be much more productive with what you're doing in order to compete. But those who do know how to compete and know how to leverage these tools are gonna kill it. And the people who have not kept up and aren't really conversant in the tools are going to drop off the bottom.
So yeah, in a couple of years, setting up API authorizations in n8n is not going to be difficult for non-technical people. Pretty soon, you'll be able to just talk to n8n and say, "Hook it up, get my credential, and get it done." But there'll be some other functionality, some new thing that's a little bit trickier to do that is not quite AI enabled yet, and we'll need to move on to that next thing. So it's turning into a tread mill. You've got to move very quickly to keep up with AI. But if you do, our world is going to be exploding with new products and services for organizations like yours to build.