Codus Review: Why AI Harnesses Are the Future of Agentic SEO

James Dooley (0:00)

Codus Review.

Today, I'm joined with Kasra Dash, and we're going to be reviewing the Codus, which is actually on the website, tricodus.com, and why I personally believe it's the best AI harness there is on the marketplace, how it enables to deploy autonomous agents and do agentic SEO. So Kasra, do you want to get started a little bit of like, what are you using Codus for?

Kasra Dash (0:30)

Yeah, so I am primarily using Codus for agentic coding. It allows me to spin up like four different variances or quadrants, if you will, of like multiple agents. Now, what a lot of people might not understand about a harness, even if you're not using one at all and you don't even want to use Codus, I just recommend using a harness, specifically because of the fact that if any of these LLMs, like for example, if Claude was to up their price 10x, you would need to stop using Claude code. And that basically just means all of your workflows, everything that you've worked on has literally just gone overnight. Same goes with like Codex, if they decided to do something, maybe they became a little bit more dumber from an LLM smartness point of view, would basically mean that you would need to stop using it.

Now, that's the number one reason why I use a harness, so I'm not actually stuck to using an LLM's harness.

James Dooley (1:33)

Yeah, for sure. I mean, I love the idea that, like you said, you can switch between Codex and also between Cloud Code, and sometimes getting one of them to quality control the other, because sometimes they come back in with different ways of doing things. Obviously, you can connect all this within one dashboard, like Higgs Field and Hagen and Google Search Console. So you're able to then start connecting like self-healing mode with SEO, or if new queries start to show up in Google Search Console, and it's a query defines a new page, you can just easily get Codex to then go and create that content. Loading existing websites in and then adding, just literally going, this is how I want the AI agents set up for Schema, go into all these pages and deploy Schema across the whole website. The amount of times that we've done that and it's binded everything together and helped out, it's brilliant for myself.

I like using, I don't know if you're doing this or not, but I love using the brain, and then I get in the brain then to deploy all the different quadrants. So it could go and deploy two or four or even eight different autonomous agents to run at the same time. For anyone that's not used Codus, can you explain a little bit about that, about the brain and how it deploys autonomous agents?

Kasra Dash (2:50)

Yeah. So let me share my screen so people can actually see, because there's a little bit more of a visual thing as opposed to us just chatting away.

So on the right-hand side here, we have got what's called the brain. So the brain will basically do all the planning and create all the tasks. And then here we have got four different quadrants.

So for example, quadrant number one might be tasked to build the design of a website. Quadrant number two might be tasked to go and grab some images or go and create some images for the website. And quadrant number three might be tasked to create a contact form. Quadrant number four might be tasked to create a topical map to help with your SEO. So all of these different quadrants are doing different tasks and the brain is basically orchestrating them all. And also as well as with all of these different quadrants, you could use different LLMs. So you can basically have Claude or Codex, for example, being the entire orchestrator of everything. And then that delegates the task to whichever LLM it thinks is the most efficient at that task. So for example, GLM might be really good at creating videos. So it's going to always dedicate the tasks for videos to GLM. So that's the difference between just using Claude Code and using an actual harness.

James Dooley (4:27)

Yeah, for sure. I mean, with regards to that and using the brain, I love the idea of it being able to use the right models. So even I prefer Claude Code. Now, I know there's a debate and some people think that Codex is better. I think whoever comes down to preferential, what they prefer, but I'm using Claude Code, but I can't get Claude Code to create my images. So then what I'm wanting to do is go into Image 2 and use ChatGPT for that or NanoBanana. All this within one dashboard can be deploying the different agents to do the different tasks. I also like the idea from Ferri, from Search Intelligence, who is the founder of Codus.