AI Agents Are Changing How We Run Our Agency
Why I Haven’t Posted in a While — And Why That’s a Good Thing
G’day Rankers, welcome back. Things are moving incredibly fast right now, which is exactly why I haven’t recorded in a little while — I’ve simply been too busy. But busy in the best possible way, and I want to bring you up to speed on everything that’s been happening.
The Hermes Agentic Harness
About a month ago — maybe longer, around the time it launched — I started using what’s called the Hermes agent, or the Hermes agentic harness. The experience has been mind-blowing. To give you a sense of just how much my workflow has shifted, I haven’t used Claude for about a week, which is highly unusual for me.
So, what does this all mean? Well, one of the guiding lights I’ve given myself as things accelerate is that anything I do with AI and these tools must directly benefit my clients — it must translate into more sales for them. That philosophy is what led us to build SKAW.
What Is SKAW?
SKAW is a tool we built specifically to answer all the questions I have about my clients’ sites. We’ve built SKAW for Shopify, we’ve now got SKAW for WordPress, and SKAW for WooCommerce is on its way. The goal is to surface the insights that matter and let us act on them faster.
When the Shopify AI Toolkit came out two or three weeks ago, it completely blew my mind. Suddenly, this agent-first approach to working on a client’s site speeds things up immeasurably. I mentioned before that a task which would have taken someone days was completed in minutes. That’s the kind of shift we’re talking about.
Building Sub-Agents with Full Client Context
What’s happened most recently is that the work we were doing with SKAW, and what SKAW was finding, led us to dig deeper with the Hermes agent. Inside Hermes, I’ve already got different sub-agents split out and dedicated to specific roles. One is a Shopify engineering agent. One is a search engineering agent. I have another agent specifically for helping me prep for client meetings.
The real beauty of these agents running on my machine through this harness is that they have full context about each client. They know about client meetings, client goals, and client roadblocks. Everything we do is framed with that knowledge. It’s like having an incredible assistant sitting right next to me, handling all of this in the background.
Of course, what comes with that is that I’m a lot busier now — building these agents, tuning them, and building out their individual skills.
A Real-World Win: Schema Fixes and the Google Shopping Bot
Each of my agents has a set of separate skills. My search engineer, for example, has a collection of Shopify-specific skills. Recently, we built a skill for a particular schema type and applied it to one client. We had noticed a drop in some of their product snippets and other metrics, so the agents diagnosed the problem, proposed a fix, and then applied it.
Before anything happens on a client site, there’s a built-in backup procedure. Everything is backed up by the agents first, and then those backups are also pushed to GitHub daily. So, the process is: diagnose, propose, backup, fix. That time lag between identifying a problem and resolving it used to be massive. Now it’s not.
What was exciting to see after we applied this schema fix was that the Google Shopping bot started coming back in SKAW. Seeing that directly on my desktop was fantastic, because it suggests there is a real, direct relationship between the fixes we’re applying and crawl behaviour improving. I want to be clear — it’s correlation at this stage. I need at least another two or three sites to properly confirm this. Interestingly, we’re not seeing this reflected in the crawl bot traffic inside Google Search Console, but we are seeing it in SKAW, which is great.
An Agency Operating System
This workflow is changing the entire way we’re going to run our business. We’re essentially building an agency operating system. And this isn’t something I’m building to sell to someone else — it’s something that’s going to make us perform better for our clients.
I know now that I have a genuine superpower when it comes to Shopify sites, because my agents that work on those sites have a bank of skills that I’ve already proven with real data across multiple clients. I’m really excited about where this is heading.
Getting Started as a Shopify Merchant
If you’re a Shopify merchant and you want to get started with any of this kind of work, the easiest path is probably going to be through Claude Cowork, or even better, Claude Code. You can add the Shopify AI Toolkit directly into Claude and start getting a feel for what it can do. It will help you manage your store through an agent, and it’s a good entry point before going deeper.
Why I Chose the Hermes Route (Model Agnosticism)
The reason we’ve gone down the Hermes path rather than relying solely on Claude Code or any single platform is that I want to be model agnostic and platform agnostic. I’d probably be using more Anthropic models like Opus 4.7 if I could access the tokens I’m already paying for — we’re spending around $250 a month between our Teams account and a personal account. But I can’t use that subscription inside Hermes. With ChatGPT and a $20 account, I can. And it just so happens that GPT 5.5 has been fantastic for this work.
I used Anthropic’s 4.7 early on and it was wonderful, but last week my experience with Claude Code on the desktop app was genuinely difficult. The agents weren’t working properly, there were memory issues, and the model just felt off. And it wasn’t just me — I saw a lot of other people saying the same thing online.
That’s a real business risk if you’re building something on top of these platforms. And it was reinforced by what came out of Anthropic last week — rumours that Claude Code in the future might only be available on more expensive plans. To be fair, today that changed again, and they’ve apparently doubled the compute power available across all plans. But that kind of movement — platforms shifting their rules and pricing quickly — is exactly why I want the flexibility to take the best model from whoever has it at any given time, plug it into my own agent setup, and not be locked into any one company’s agentic harness.
The beauty of building agents the way we’re building them is that they are transferable. Everyone else in the office is currently using Claude Code and the agents we’ve already built for them in that environment. I’m the only one using Hermes right now. But from what I’ve seen, having now used it for however long it’s been available, there’s no going back for me. I don’t know exactly what AGI looks like, but I think it looks a lot like Hermes combined with GPT 5.5.
Social Media and AI-Generated Content
One more thing worth mentioning — pretty much all my social media is now being handled by Hermes. I’d encourage you to critique it if it sounds too AI-ish, because it probably is right now and I’m actively working on improving that. I don’t want to churn out slop. What I’m trying to do is have my social content generated from the real work that we’re doing here. I’ve got an agent that looks at what we’ve been working on, then tells that story for everyone else — without revealing any client data, of course.
And now it’s also got the ability to make short videos. The one it made this morning was generated using something called Hyperframes from HeyGen, which you can also use inside Claude Code and Claude Cowork. I didn’t give it any real direction — I just said make me a video to accompany a post, and that’s what it produced. There are a few layout issues but given that I gave it almost no instruction and it has full context of what we’re working on, the fact that it can now just start creating this kind of content is genuinely impressive.
Wrapping Up
It’s all a bit clunky right now, but this is the worst it’s going to get. So, get on it. Start using the tools, start exploring, and if you have any questions, drop them in the comments or find me on LinkedIn, YouTube, X, or Facebook — wherever Hermes has posted this for you to find. If you want to reach out and have a chat, you can email me at [email protected]. Hopefully that’s been helpful, and we’ll see you next week. Thanks very much.
Jim’s been here for a while, you know who he is.
