I build production systems with AI. Not by writing code — by seeing what needs to exist and figuring out how to make it real. This is how I think about that.
I started working with AI in late 2022, back when most people still thought it was a novelty. I wasn't chasing a trend. I was stuck. Too much drag in every system I touched, too much friction between what I could see needed to exist and what I could actually build. AI removed the bottleneck.
I've never been an engineer. Never called myself a developer. But over the past two years I've built 20+ production systems — automated pipelines, live dashboards, multi-agent research workflows — that run on schedules and do real work while I sleep.
People keep asking me to explain how that's possible. This is my attempt.
That's the part nobody talks about. Building with AI doesn't feel like programming. It feels like a creative act — closer to drawing or cooking or playing an instrument than writing code.
AI lets me reach in and pull that out. Not by translating the vision into a programming language — which always felt like compressing a 3D idea into a flat instruction set — but by describing what I need and iterating until the system responds to my intent.
There's a vision that exists fully formed: the dashboard someone opens at 7am, the research wire that surfaces what matters before anyone else sees it, the system that runs while you sleep and has the answer ready when you wake up.
When it's working, it's flow. The kind where hours disappear. When it's not, it's the frustration of an instrument going out of tune mid-performance. That distinction matters. The frustration is aesthetic, not technical. I don't care that the JSON is malformed. I care that the thing I can see in my head isn't showing up in the world yet.
I call myself a vibe coder. But I debug webhook payloads at 2am. I've diagnosed context pollution in LLM conversation states. I've built parallel pipeline architectures that process dozens of data streams into formatted outputs.
The gap isn't between what I say and what I do. It's in the vocabulary. "Vibe coder" is a rejection of the engineering identity — the culture, the credentialism, the gatekeeping. It's not a rejection of the work.
I think the tension is actually resolved, and the resolution is the interesting part: I build better because I don't think like an engineer. An engineer would over-architect. I start with the desired output and work backward. "I need this to run at 7am and produce a briefing." Then I solve whatever's in the way. That's closer to how a chef develops a dish than how a developer ships a feature.
The word I've landed on is architect. Not because it sounds impressive. Because it's accurate. I design systems. I know what goes where, what each component is for, how they talk to each other. I just don't write every line from scratch. And increasingly, that distinction matters less.
A few things I've learned that feel load-bearing:
Which AI you use matters less than how you orchestrate it. A GPT wrapper is putting a new face on ChatGPT. What I build is plumbing — data sources, pipes, filters, with AI as one valve in a system I designed.
Traditional programming languages are a degraded translation of intent. Natural language is higher-fidelity. I think the future isn't text-to-code — it's post-code, where the system understands what you mean without needing it explained line by line.
Execution is increasingly automatable. Knowing what should exist — and what it should feel like to use — that's the hard part. That's always been the hard part. AI just made it visible.
Small ask, tight feedback, high signal. I think in versions, not one-offs. Fast loops compound. Perfect plans don't survive contact with reality.
Most expertise never gets documented. Publications capture findings — they don't capture how someone thinks. The instincts, the pattern recognition, the failed approaches. I build systems to keep that.
Here's what I think is actually happening, underneath all of it.
Most institutional structures — companies, agencies, teams — are optimized for process and compliance. They systematically filter out non-linear thinking. Associative leaps. Intuition-driven pattern recognition. The kind of skip-thinking that certain minds run on.
AI doesn't just let small teams compete with big ones. It removes the structural barriers that have always prevented certain kinds of minds from building at scale.
People who think in narratives instead of specifications. People who see the whole system before they see the parts. People who couldn't ship because the translation layer between their vision and the machine was too thick.
That layer is dissolving. And when it does, it turns out some of the best builders were never in engineering. They were stuck on the other side of a wall that doesn't exist anymore.
I'm not a tech bro. No hustle culture, no breathless hype about the latest model drop.
I'm not a doomer. I don't think AI is going to end us. I think it's going to be wielded — well by some, badly by most.
I'm not performative. I don't post for engagement. I don't demo without production behind it. Everything I've built runs unsupervised. It does real work. The systems exist whether or not anyone's watching.
I'm not trying to build an audience. I'm trying to build systems. If people find that interesting, great. If not, the work still matters.
I know the terrain. Data leakage, false outputs, weaponized attention, collapsing boundaries. I've internalized the ethical ambiguities.
It's easy to be frozen by all of that. To theorize endlessly, to wait for the perfect framework before you touch anything. I've been there. But at some point I decided that starting imperfectly beats waiting for permission that never comes.
So I move. I build the thing, see what breaks, fix it, build the next one. The understanding comes from doing, not from standing at a safe distance and having opinions.