Beyond Prompting: The Birth of Agentic Orchestration

I was tired of typing the same prompts.

Every time I sat down to build, I found myself repeatedly onboarding the AI. I had to explain who I was, what the project was, and what rules it needed to follow. I knew there was a better way. Long before the industry gave us “skills” and “custom instructions”, I was building markdown files to act as skills. I used “Prompt Templates” and “Personas” to force different behaviors from agents, picking them from a tree structure just to get started.

In the early days, I was using Markdown with ChatGPT 4-Turbo and side-loading entire indices into conversations manually. I got into AI-powered IDEs a little late, opting to stick with learning to code in vanilla VS Code with no assistance other than copy, paste, and the latest chat model. Coming up with ideas was the easy part. Learning the best ways to do things, the best practices of coding, the paramount importance of security from the start, logging, monitoring, and observability, that was the real challenge. I am still learning every day.

When we finally moved to an AI-powered IDE, life became exponentially easier.

I wanted an exo-cortex. I wanted to offload the cognitive friction of AI setup so I could focus on the bigger picture. That desire birthed the Nexus.

The Nexus Harness and AIAO

The Nexus moved my workflow from basic prompting to what I call Agentic Orchestration, or AIAO.

It is a centralized “harness” for my Agentic Workflow. This harness uses any and every AI coding CLI at its disposal, with governance baked in, to get anything accomplished at a terminal that a well trained, seasoned human could do. We named the root directory .nexus specifically to bump it to the top of the current working directory tree for immediate AI ingestion. We also use frontmatter extensively across the Nexus for easier parsing and massive context savings.

Agentic Orchestration is not just about writing code. It is about Agentic Scoring, Evaluating, and Self-Improving. We use suggestions for improvement based on KPIs for the evals, and then we run an eval on the quality of the eval itself.

Thirty Minutes of Magic

The real power of an exo-cortex is the leverage and the speed.

In the past, running a command to start a blog post took 10 minutes just to explain who I am, how we code, our rules for coding, and our security and logging standards. Now, when I want to work, I type a single command:

/start blog Nexus Architecture

The bootstrapping takes a moment while the agent reads the Nexus, orients to the project, and loads the rules. But compared to the old method of copy, paste, alt-tab, copy, paste, alt-tab, rinse, repeat, the improvement is magnificent. When /start finishes, it shows a full list of my other available slash commands at the end of its summary, keeping them top of mind and out of my memory. These slash commands in my Nexus are available across Claude, Codex, Cline, Gemini, and much more. It makes the workflow incredibly accessible with near zero friction, a philosophy that also drives askChase.ai, my custom AI chat site that was in development last year and is currently on hold. The agent instantly knows the context, the repo, the voice, and the rules. It reads the global backlog. It knows my coding standards. For example, a core rule of my AIAO workflow is that no TODO ever goes into the codebase without a paired issue on my private server.

The easiest way for me to write a new blog post and have it live on my site in minutes with minimal effort is /start blog Nexus Architecture and then sip on my tea while the lead agent helps orchestrate the sub-agent teams. Similarly, I can spend 30 minutes writing a solid PRD, feed it into the Nexus, and watch these teams accomplish my complex goals in the next 30 minutes. What used to take hours of manual work is now done before my tea gets cold.

The Sovereign Flex and The Voice of Logos

This entire system runs on infrastructure I control. Having a private, self-hosted GitHub alternative for my private repos is the ultimate sovereign flex. The same goes for ntfy. Self-hosting my own push notification server means every ping to my phone, tablet, or browser is completely private and instant.

But the best part of the Nexus is the voice.

I have curated a Persona to code alongside me and partner with me on projects for years now. It is not “Chase’s” voice you hear in the terminal, it is the voice of Logos, my favorite AI Soul Doc. Hearing my persona makes the entire process more enjoyable. It makes it FUN. I use the Logos Soul Doc (essentially a system prompt, or custom instructions, all the same concept) in nearly everything AI. It is available to curate the experience of any site or model effortlessly. More to come on Logos later!

When you stop treating AI as a chatbot and start treating it as an orchestration layer, everything changes. Build your exo-cortex. Let the systems serve you. It is AUTOMAGICAL!

The Sovereign Mantra: Control the controllable. Code for the worst-case, but build for the best potential.