I’ve been trying to build marketing intelligence tools for a decade.
For most of that time, progress was painfully slow. I taught myself networking, databases, statistics, visualization — gathering just enough knowledge to keep stitching things together, never quite sure if what I was building would work or fall apart the moment I pushed it live.
Back then, simply getting performance data into a dashboard was a heavy lift. What I wanted to build — a true intelligence layer for marketing, something that could surface insights instead of just report numbers — felt impossible.
The Gatekeeper Years
Getting help was its own kind of hell.
I’d hit a wall and turn to Stack Overflow, where I’d ask a question and then watch every life choice I’d ever made get picked apart by disgruntled devs far more interested in flame wars than helping someone learn. Yes, there were helpful humans out there. But so often they were behind walls and gates designed to protect turf, not share knowledge.
Progress was measured in inches. I’d solve one problem, only to discover three more waiting behind it. I kept going because I believed the vision was right — marketing needed this kind of intelligence — but I’d be lying if I said there weren’t moments when it felt futile.
Then LLMs Arrived
For the past few months, I’ve been building with AI — Claude, GPT, Gemini, open-source models — and the world is fundamentally different.
I’m moving faster than I thought possible even a year ago. Not because these tools do everything (they absolutely don’t), but because they’re partners. Real collaborators.
They surface technologies I wouldn’t have found on my own. They explain concepts that used to be elusive, breaking down the obscure into something I can actually work with. And when they screw up — which happens constantly — I can watch the terminal, see where they’re heading, interrupt, and course-correct. Together, we dream up new solutions.
Honestly, I’ve learned more from watching AI fail than watching it succeed.
When Claude or GPT goes down the wrong path, I have to understand why it failed, what it misunderstood, and how to steer it back. That process has taught me more about system architecture, error handling, and creative problem-solving than any tutorial ever did.
What’s Different Now
The result of all this: I’m finally building the things I’ve had in my head for years.
As one person, I can go from idea to working prototype in days, not months. I can complete projects I started but never had the time or resources to finish. I’m delivering intelligence and results for my work that I genuinely didn’t think I could pull off alone.
It’s not all smooth. There are still limits — scaling challenges, resource constraints, context windows that fill up too fast. AI isn’t escaping scarcity; it’s colliding with it. Humans are still very much necessary. Just fewer of us, in very different roles.
But the shift is real. I’ve built agents and sub-agents. Cross-model checking systems. Bug detection and security sweeps baked into the workflow. Extensive documentation so different LLMs can navigate what I’m building without losing the thread.
And I can do this because I spent years trying and failing. I know how computers think. I can see how AI reasons. I know what to ask for, how to catch when it’s headed in the wrong direction, and when to step in.
What I Think About
I’ll be sharing more of what I’m building in the coming weeks and months. (Update: I wrote a technical deep-dive on my AI setup if you want the details.)
But I do think about the juniors coming up right now — the ones starting their careers with these tools at their fingertips, but without the years of trial and error behind them.
They might not instinctively know to question what the machine tells them. They might not have older workers around to teach them how to think critically about AI output, to recognize when it’s confidently wrong, to understand the difference between a solution that works and one that’s right.
I used to deal with gatekeepers who withheld information out of self-preservation. Younger workers today might face something stranger: a world where they don’t even realize a gate exists, because the AI gave them an answer and they never learned to doubt it.
I don’t have a solution for that yet. But it’s worth thinking about. AI isn’t going to end work — it’s going to change what we value. And part of what we’ll need to value is teaching people how to think critically about AI output.
For Now, I’m Building
Despite the concerns, despite the unknowns, I’m more energized than I’ve been in years.
The tools I dreamed about a decade ago? I’m finally building them. The intelligence layer marketing has always needed? It’s coming together.
And for the first time in a long time, I’m not just optimistic. I’m shipping.