AI didn't make me lazy. It handed me my own subconscious.

Right now I’m eating turkey pepper soup at my work desk, and I have multiple jobs running with AI.

One agent is taking a client project from brief to restructured profile to redesign, start to finish. First time I’ve ever let it run the whole thing. I don’t know if it will work. Another is deploying new functionality to one of my platforms. And in Obsidian, I’m voice-dumping to a third one, playing with an idea I want to post tonight. The post you’re reading now.

For the past year I’ve been all the way into AI. Like, REALLY into it. Last year, I was very skeptical. This year I had my road to Damascus moment, and I wrote about what I saw back in April, after building eight apps in six weeks without knowing how to code.

I said work was no longer about doing the work but directing it. That you need a meta approach: understand the work, its value, the systems and workflows around it, not just the task in front of you. I said AI was a motorbike for the mind, but it still needs a mind to steer it.

It’s hard to explain sometimes, but something in my mind shifted. I felt the shape of something, innately, and I’ve been chasing it ever since. The months since have been me living inside that idea. Pushing it to new and strange places.

Hold on. My agent just sent back design references. I’m laughing. Goddamn, that actually looks pretty good. Okay. Wow.

Where was I...

Oh yes, I’m always thinking about the work, not just the task.

What does it actually take to do the thing? What are the systems around it, what is the workflow, how does the work get done? Because once you can see that clearly, then you can abstract it. And once you can abstract it, you can encode it. And that is a profound thing.

If you can encode the work then you can manipulate it, engage with it in different modes. You can switch things around. You can add force multipliers at different points. You don’t just know what you are doing, you have a sense of how, and it can live outside your head and talk back to you.

I’ve spent months encoding ideas into software. I built a brand operating platform that walks you from audit all the way to growth. It’s built on my methodology, prompts, sequence, step by step. I found the methodology as I built the different parts, intuitively, in the dark it seemed half the time, with moments of radiant brilliant recognition.

That’s an idea encoded into a framework. It works.

Earlier this week I wrote a book about that methodology in five minutes. Assembled it, really. Five minutes of assembly, sitting on a vault full of notes, a pile of blog posts, months of conversations with my agents, and ten years of doing the work. First the idea lived in chat. Then it became software. Then content. Then the book. And halfway through restructuring the book I realised part of my software was wrong, and now I have to go fix it. That is what it looks like to attack an idea from every side.

Each medium teaching you something new about the whole thing.

Moving at this speed poses some questions. After months of it, “can I do it” is not the question anymore. It’s more like...okay, where are my leverage points? How do I stay sharp, hone my true unfair advantage, and not succumb to outsourcing the real thing to the machine?

My intuition points me to the loops. With these tools, the loops are faster. Iteration, feedback, build, review. That is the defining feature. When you can move from idea to software to content to book to corrected software in a week, what would take months, decades even, to surface and refine can come out much faster. The sheer speed of the loop makes you better faster...I think...It is a lot. The brain is still adjusting.

Now on to design...taste. That has always been the tricky part. That’s the piece I wrote off in April as the untemplatable part. And even as I push further into AI (I’ve moved from working in Claude chat, to cowork, to code, to terminal, and now my own agentic dashboard, in months), lately I’ve been circling back to doing more design, more writing, more directing. I have a clearer sense of what can be reliably delegated and what can’t. For now.

I’ve been seeing a lot of things as code lately. By that I mean breaking everything into its component parts and seeing how they relate, how they combine. I have been thinking about brand design the same way.

Here’s the workflow I’ve been running for a few weeks. Point the AI at a reference, or an old document, and have it extract everything, the ideas, the layouts, the type, the colour, the textures, the imagery, into a structured brand guide template that I can parse and it can understand.

Then we work. Sometimes twelve, thirteen, fourteen hours straight on one guide. Tweaking, tweaking, tweaking. I still work very hard. More now, I think. But every time I work hard now, the payoff is more than the thing. The payoff is the lesson. The payoff is me understanding my own process and learning how to communicate it to the machine. The work deposits something now, where it used to just get lost to the subconscious.

People say AI makes you lazy. I don’t know, not really my experience man. I think it is the most incredible extraction tool for everything you’ve been doing for years, for better or worse. All that subconscious knowledge, the stuff you were just good at without knowing why, for the first time you can extract it, name it, use it, tweak it. Being good at it becomes knowing why you’re good at it, becomes being able to hand it over. Conscious competency, then conscious delegation.

And once you have it named, and in your hand, it is a very powerful tool.

When you know the pieces, you know what to tweak. I understand the visual side of branding differently now. A component is no longer there because I like it. It is there because it makes sense for what the brand is supposed to be. Every tweak points at the concept. Naming my own process made me better at the process. The extraction improves the thing it extracts from.

Which brings me to tonight’s test. I’m tired. It’s been a heavy week, two days straight deep in platform UI, and a client needs this thing. And I’ve been wrestling my way to a simple admission: it is time to let go more. To embrace the leverage I’ve been building.

Sometimes you hold the AI’s hand all the way, step by step. But after you’ve done that long enough, like training a junior, there comes a day when you say: just do this. And what comes back is decent.

Tonight is that day. That’s the agent running the client project while I chop pepper soup.

And maybe there’s one more step. Maybe the agent that knows me, that has my context, my workflow, my biases, all the things in my head I usually have to connect myself, maybe at the end of the day I can jam with it the way I always tear the workday down, and out of that comes a blog post. In my voice. Edited by me. Fueled by my flow, by what happened today.

Tonight is the first time I’m fully embracing that as a valid part of my process. Dump to draft to posted, tonight. Because honestly, I’ve been having ideas and documenting concepts and lessons faster than I can deploy them, and I’ve been deploying fast.

So this is another experiment: what happens when I post the ideas as I have them, instead of sitting on them until they’re perfect. I want to join the conversation. I want to hear what you’re learning and trying, because some of what I’ve found is genuinely interesting, and I suspect some of what you’ve found is too.

That’s what this is. The same creative flow. With the brush, with the mouse, with the pen.

I am pouring my heart and mind out, the way I always did with a piece of paper in front of me. The way I always did with the blank Notion page.

The surface just thinks now.

The tools got smarter.

The hand is still mine.