Wrestling with Agentic AI
AI Strategy

Wrestling with agentic AI, Part 1

June 15, 2026 Agentic AI AI Adoption AI Strategy

"I feel like I'm falling behind, but I don't even know what I'm behind on."

Maybe you've said it, or heard it from a friend. It's everywhere.

That's the thing about AI. The tools really aren't that hard to use. But the way people talk about them. The words are just for people who already get it. Like there's an inner circle, and until you're in it, you're just on the outside looking in.

I'm part of a group working at the sharp end of AI. People who are building, implementing, and helping others figure it out. We're working at the edge of what's actually possible. It's baffling.

Not because the concepts are over my head, but because the language assumes you already know.

I still feel like I'm on the outside and I want in.

I keep seeing something I can't unsee. The idea of a system that runs on its own while I sleep.

I could keep using AI the way most people use it. ChatGPT for drafts, Claude for thinking, automation stitched together where it makes sense. But I keep seeing something I can't unsee. The idea of a system that runs on its own while I sleep. One that gets smarter about what I need over time. That's the real edge of what AI is. And it's also the edge of what I understand.

I really want to get it. So I'm trying full-tilt.

My agent runs on a system called Hermes. The agent inside it, the one who shows up on Telegram, is Joe. He's not a chatbot. He's my AI chief of staff. He runs on Railway, talks to me through Telegram, uses Claude as his brain. The sheer number of systems makes my head want to explode. Why can't this be easier??

Every morning he is supposed to pull my calendar, check the news, grab the Halifax weather, and get me up to speed before I'm caffeinated. He knows what matters to me because of a tool called Honcho, which gives him semantic memory. Over time he learns what I actually want to know, and things are supposed to get better and easier.

On a good morning, I wake up and the hard work of figuring out my day is already done. That's the whole point.

My first working version was giving me the time in UTC, so my 9am looked like 1pm. Nothing made sense. Fixed it and felt pretty good about myself for all of 20 minutes. Then I found out Joe had been making up the weather for two days straight because he had no internet access. I only figured that one out when the weather had nothing to do with what was happening outside. Two days. This has got to be easier.

It's frustrating and I'm still wrestling with this. If you see me around town with a few less hairs, you'll know why.

Come along for the ride. I'll keep writing about this as it hopefully comes together.

Frequently asked questions

What is agentic AI?

Agentic AI refers to AI systems that take autonomous action on a person's behalf rather than simply responding to prompts. An agentic system like Joe, Carol Roderick's AI chief of staff, gathers information, makes decisions about what matters, and surfaces what's relevant without a human directing each step. This is different from standard AI assistants, which require a human to initiate every interaction.

What is semantic memory in an AI agent?

Semantic memory in an AI agent is the ability to retain and build on information across conversations over time. Carol Roderick uses a tool called Honcho to give Joe semantic memory, which means Joe learns what Carol flags as important, what she skips, and what kind of information is actually useful to her morning briefing. The result is an agent that improves at its job the longer it runs.

What is the difference between AI automation and agentic AI?

Automation executes a task that a human has defined in advance. Agentic AI exercises a form of judgment: it decides what to do, how to do it, and what to surface based on context and learned preferences. Automation is something you trigger or start, and agentic AI is something that initiates on its own. The difference is the difference between a tool and something closer to a colleague.

Will you be writing more about this?

Yes, of course. I'm obsessed with figuring this out right now, so there will definitely be more. You'll get my wins, my fumbles, and what it all means for how AI is going to change the way we work.

Carol Roderick, PhD is an AI Integration Strategist for expertise-led businesses in Halifax, Nova Scotia. She is an approved delivery partner for the Digital Nova Scotia AI Digital Adoption Program, which covers up to 75% of qualifying work for eligible NS businesses. Get your AI Reality Check, a free assessment of how you're responding to the pressure to adopt AI, at carolroderick.ca.