The plumbers of the digital world.

Everyone's being told to implement AI right now. Every conference, every LinkedIn post, every CEO email. "Implement AI. Use AI. Become an AI-first company." And almost nobody knows what that actually means. They paste a prompt into ChatGPT, paste the answer into an email, call it adoption, and wonder why nothing's changing.

That's not implementing AI. That's using a chatbot. There's a difference the size of a house.

What people think AI implementation is

Open ChatGPT. Ask it something. Copy the response. Paste it into your actual work. Repeat. Maybe graduate to Claude when you run out of GPT quota. Maybe try Perplexity when you want something cited. Now you've got three tabs open, you're context-switching between them, and the work is still getting done by you. The AI is a better Google. That's it. You've added a tool to your toolbox. You haven't changed anything about how the work gets done.

I'm not mad at people for this. Honestly, it's what the tools teach you to do. Every onboarding is "here's a chat window, type your question." That framing is the problem. It says the AI is an assistant you talk to. It's not. At least not where the actual change lives.

What it actually is

Real AI implementation is plumbing. It's what happens between the tools, not inside any one of them. It's connecting your email to your CRM to your calendar to your document vault to your finance system to your outreach pipeline to the AI agent that reads all of it and decides what needs to happen next. The AI isn't the product. The AI is the engine. The product is the system of pipes that feeds it work and pulls results back out.

Almost nobody is doing that. Because it's hard, it's invisible from the outside, and it doesn't look like a demo. There's no screenshot that captures "my CRM now triggers a qualification flow that updates my calendar when a lead responds to an email draft that was generated from my own voice style guide." That's seven systems talking to each other. No single vendor ships that. You have to wire it yourself. And wiring is where the real work has always lived, in any technology wave, going back to electricity.

The plumbers win

We think we're the kings of the planet. We built something that can do everything we do digitally. Read, write, research, decide, draft, follow up, schedule, remember. And collectively we're still treating it like a search bar. The people who figure out the wiring are going to eat. Not the people who build the AI. Not the people who sell the AI. The people who can plumb it into a real business, end to end, and make the pipes invisible.

That's the opportunity. That's what Obsidian AI Labs is. We're plumbers. We show up, we look at your stack, we connect the systems that should have been connected five years ago, and we give you something that runs while you're sleeping instead of something you have to babysit every morning. It's not glamorous. It's not a demo. But it's the actual change.

What I'm actually doing right now

I have three agents running simultaneously building out my entire infrastructure. Not one. Three. A primary orchestrator on my desktop. A dedicated sales agent on a droplet. A dedicated finance agent on another droplet. They're each running their own workflows, executing their own tasks, reporting back. They set a pace that is outlandishly crazy to try to keep up with.

It has been physically taxing and mentally draining. I'm not going to pretend otherwise. Watching three systems move in parallel, each one doing work that would take a person a full day, while trying to keep the whole thing coherent, is a new kind of exhaustion. But I think I'm getting closer to having everything I need done. And the pace they're setting is real. What would have taken me six months to build manually, I've built in three weeks because I'm not the one typing.

Let go of control

Sometimes you just have to let go of control. That's the part nobody talks about, because it's not a technical skill, it's an emotional one. When your sales agent is drafting an email you didn't write, in a voice that's very close to yours but not exactly yours, you have a choice. You can edit every line and make it exactly how you'd say it, which takes you back to doing the work yourself. Or you can let it go, read it for correctness, and hit send. The second one is harder. The second one is the whole game.

I came from cybersecurity. Nine years of hands-on technical work. Control is baked into how I think. Verifying, auditing, double-checking, doing it myself because that's the only way to know it was done right. That instinct is useful. It's also the exact thing that was keeping me small. I had to learn the difference between "the system needs a guardrail here" and "I just don't trust it." One of those is engineering. The other is ego.

Stop trying. Start doing.

I want to build this for everyone. I want the world to stop trying and start doing. Most people I talk to are three months into "thinking about implementing AI" and zero months into having any of it actually running. The gap between the people who are going to benefit from this wave and the people who are going to watch it pass them is going to be measured in months, not years. And the dividing line is execution, not intelligence.

This blog is part of how I'm thinking out loud about it. I'll be using it to jot down where I am, where I'm stuck, what I'm figuring out. It's public. Everyone can see it. I don't have a fully formed plan for the format yet. What I have is a working infrastructure, a lot of live battle scars, and the conviction that the world needs to rethink what implementing AI actually means. It's not a chatbot. It's not a tool. It's a new way of working that nobody's shown you yet.

Be the plumber. Or hire one. But stop pretending the chat window is the whole thing.

← Back to all posts